Authors: Al Kari
Abstract: The proliferation of Large Language Model (LLM) architectures presents a fundamental challenge: valuable, task-specific behaviors learned through fine-tuning methods like Low-Rank Adaptation (LoRA) are effectively trapped within their source model's architecture, herein referred to architectural lock-in. Existing transfer methods attempt to bridge this gap by aligning the static weight spaces of models, a brittle and indirect approach that relies on tenuous correlations between parameter geometries. This paper introduces a fundamentally different and more direct paradigm: the Cartridge Activation Space Transfer (CAST), a novel framework that liberates LoRA-encoded behaviors by learning a direct, nonlinear mapping between the activation manifolds, the geometric structures formed by the model's internal neuron activations, of two distinct LLM architectures. CAST treats a pre-trained LoRA as a frozen "behavioral kernel." It learns a set of lightweight, bidirectional projection heads that translate the target model's activation stream into the source model's latent space, apply the frozen kernel, and project the result back. This process, trained on a general text corpus without any task-specific data, effectively decouples the learned skill from the source architecture. We demonstrate that CAST enables true "zero-shot" translation of any standard LoRA adapter. Our experiments, including transfers between heterogeneous model families like Llama-2 and Mistral, show that CAST-translated adapters achieve 85-95\% of the performance of a LoRA fully retrained on the target model, quantitatively outperforming current weight-space transfer techniques and establishing a new state-of-the-art in model interoperability.
Authors: Zhiming Lin
Abstract: Multi turn intent understanding is central to task oriented chatbots, yet real deployments face tight token budgets and noisy contexts, and most retrieval pipelines emphasize relevance while overlooking set level diversity and confounds such as more context or exemplar order. We ask whether retrieval diversity, rather than longer prompts, systematically improves LLM intent understanding under fixed budgets. We present a diversity aware retrieval framework that selects in context exemplars to balance intent coverage and linguistic variety, and integrates this selection with standard LLM decoders; the evaluation enforces budget matched prompts and randomized positions, and includes sensitivity analyses over exemplar count, diversity strength, and backbone size. On MultiWOZ 2.4 and SGD, the approach achieves strong gains in Joint Goal Accuracy under equal token budgets, surpassing strong LLM/DST baselines, with consistent improvements across K from 4 to 7 and moderate latency. Overall, the study isolates and validates the impact of content diversity in retrieval and offers a simple, deployable selection principle for building accurate, budget constrained multi turn intent systems.
Authors: Abhigya Verma, Seganrasan Subramanian, Nandhakumar Kandasamy, Naman Gupta
Abstract: Large language models (LLMs) are increasingly deployed as agents, expected to decompose goals, invoke tools, and verify results in dynamic environments. Realizing these capabilities requires access to agentic data- structured interaction records that couple user intents with tool specifications, argument-grounded calls, and verifiable execution traces. However, collecting such data from human annotators is costly, time-consuming, and difficult to scale. We present a unified framework for synthesizing agentic data using only LLMs, without any human-in-the-loop supervision. This framework decomposes generation into modular pipelines that produce complete interaction records spanning task specifications, tool definitions, policy pseudocode, natural language exchanges, and execution traces. Records conform to strict syntactic and semantic constraints, ensuring machine-parseability and faithful alignment across inputs, outputs, and tool calls. Beyond single tasks, there is support for both multi-task and multi-turn agent interactions, enabling the construction of datasets that reflect the full spectrum of tool-use competencies. To ensure quality and consistency, the framework integrates constrained generation formats, JSON-schema validation, and judge-based filtering. This paper formalizes the schema for agentic records, details the prompt design principles that guide generation, and introduces scalable pipelines for high-quality synthetic data. By providing a reproducible, LLM-only alternative to manual collection, hence advancing the development of agentic LLMs capable of robust tool use.
Authors: Zhenyu Bi, Meng Lu, Yang Li, Swastik Roy, Weijie Guan, Morteza Ziyadi, Xuan Wang
Abstract: Large Language Models (LLMs) have shown remarkable reasoning capabilities in mathematical and scientific tasks. To enhance complex reasoning, multi-agent systems have been proposed to harness the collective intelligence of LLM agents. However, existing collaboration structures are either predefined or rely on majority voting or round-table debates, which can suppress correct but less dominant agent contributions. Recent approaches model multi-agent systems as graph networks but optimize purely for agent performance, neglecting the quality of interactions. We hypothesize that effective agent communication is crucial for multi-agent reasoning and that debating quality plays a significant role. To address this, we propose $\ours$, a multi-agent verbal reinforcement learning algorithm that dynamically constructs and refines multi-agent collaboration structures. Our method defines action spaces and a feedback mechanism that evaluates communication robustness and coherence throughout the debate. The final decision is achieved through a majority vote over all the agents. We assess $\ours$ on various reasoning tasks, including mathematical reasoning, creative writing, scientific reasoning, and numerical sorting. Results demonstrate that our approach significantly outperforms single-agent prompting methods and state-of-the-art multi-agent frameworks on diverse tasks.
Authors: Alexander Boldachev
Abstract: A formalization of a subject-event ontology is proposed for modeling complex dynamic systems without reliance on global time. Key principles: (1) event as an act of fixation - a subject discerns and fixes changes according to models (conceptual templates) available to them; (2) causal order via happens-before - the order of events is defined by explicit dependencies, not timestamps; (3) making the ontology executable via a declarative dataflow mechanism, ensuring determinism; (4) models as epistemic filters - a subject can only fix what falls under its known concepts and properties; (5) presumption of truth - the declarative content of an event is available for computation from the moment of fixation, without external verification. The formalization includes nine axioms (A1-A9), ensuring the correctness of executable ontologies: monotonicity of history (I1), acyclicity of causality (I2), traceability (I3). Special attention is given to the model-based approach (A9): event validation via schemas, actor authorization, automatic construction of causal chains (W3) without global time. Practical applicability is demonstrated on the boldsea system - a workflow engine for executable ontologies, where the theoretical constructs are implemented in BSL (Boldsea Semantic Language). The formalization is applicable to distributed systems, microservice architectures, DLT platforms, and multiperspectivity scenarios (conflicting facts from different subjects).
Authors: Joong Ho Choi, Jiayang Zhao, Jeel Shah, Ritvika Sonawane, Vedant Singh, Avani Appalla, Will Flanagan, Filipe Condessa
Abstract: Large Language Models (LLMs) deliver powerful reasoning and generation capabilities but incur substantial run-time costs when operating in agentic workflows that chain together lengthy prompts and process rich data streams. We introduce CompactPrompt, an end-to-end pipeline that merges hard prompt compression with lightweight file-level data compression. CompactPrompt first prunes low-information tokens from prompts using self-information scoring and dependency-based phrase grouping. In parallel, it applies n-gram abbreviation to recurrent textual patterns in attached documents and uniform quantization to numerical columns, yielding compact yet semantically faithful representations. Integrated into standard LLM agents, CompactPrompt reduces total token usage and inference cost by up to 60% on benchmark dataset like TAT-QA and FinQA, while preserving output quality (Results in less than 5% accuracy drop for Claude-3.5-Sonnet, and GPT-4.1-Mini) CompactPrompt helps visualize real-time compression decisions and quantify cost-performance trade-offs, laying the groundwork for leaner generative AI pipelines.
Authors: Daniel Israel, Tian Jin, Ellie Cheng, Guy Van den Broeck, Aditya Grover, Suvinay Subramanian, Michael Carbin
Abstract: A central challenge in large language model inference is the trade-off between generation speed and output quality. Autoregressive models produce high-quality text but generate tokens sequentially. Diffusion models can generate tokens in parallel but often need many iterations to match the same quality. We propose planned diffusion, a hybrid method that combines the strengths of both paradigms. Planned diffusion works in two stages: first, the model creates a short autoregressive plan that breaks the output into smaller, independent spans. Second, the model generates these spans simultaneously using diffusion. This approach expands the speed-quality Pareto frontier and provides a practical path to faster, high-quality text generation. On AlpacaEval, a suite of 805 instruction-following prompts, planned diffusion achieves Pareto-optimal trade-off between quality and latency, achieving 1.27x to 1.81x speedup over autoregressive generation with only 0.87\% to 5.4\% drop in win rate, respectively. Our sensitivity analysis shows that the planning mechanism of planned diffusion is minimal and reliable, and simple runtime knobs exist to provide flexible control of the quality-latency trade-off.
Authors: Nikhil Verma, Manasa Bharadwaj, Wonjun Jang, Harmanpreet Singh, Yixiao Wang, Homa Fashandi, Chul Lee
Abstract: Large Language Models (LLMs) have redefined complex task automation with exceptional generalization capabilities. Despite these advancements, state-of-the-art methods rely on single-strategy prompting, missing the synergy of diverse reasoning approaches. No single strategy excels universally, highlighting the need for frameworks that fuse strategies to maximize performance and ensure robustness. We introduce the Select, Mix, and ReinvenT (SMaRT) framework, an innovative strategy fusion approach designed to overcome this constraint by creating balanced and efficient solutions through the seamless integration of diverse reasoning strategies. Unlike existing methods, which employ LLMs merely as evaluators, SMaRT uses them as intelligent integrators, unlocking the "best of all worlds" across tasks. Extensive empirical evaluations across benchmarks in reasoning, planning, and sequential decision-making highlight the robustness and adaptability of SMaRT. The framework consistently outperforms state-of-the-art baselines in solution quality, constraint adherence, and performance metrics. This work redefines LLM-driven decision-making by pioneering a new paradigm in cross-strategy calibration, unlocking superior outcomes for reasoning systems and advancing the boundaries of self-refining methodologies.
Authors: Soheil Abbasloo
Abstract: What does it truly mean for a language model to "reason"? Most current evaluations and benchmarks reward models' correct standalone answers--but correctness alone reveals little about the process that produced them. In this work, we explore a different perspective: reasoning is not a static chain of steps, but a dynamic trajectory where ideas interact, clash, and evolve into deeper insights. To capture this dynamic, we draw on a well-established philosophical tradition: \textit{dialectics}, where reasoning unfolds through thesis, antithesis, and synthesis. Building on this, we present SIEV, a structured framework that evaluates reasoning of LLMs through dialectics. Unlike conventional evaluations, SIEV assesses not only the conclusion a model reaches, but how it gets there: its ability to resolve tension, integrate distinct ideas, and synthesize higher-order reasoning. This lens uncovers significant reasoning gaps in state-of-the-art models even under saturated benchmarks like GSM and MMLU. For instance, GPT-5-chat, a recent model, loses over 40 points (out of 100) when evaluated with SIEV on GSM. Our findings highlight that adopting a process-oriented, philosophically grounded approach enables a deeper, more rigorous, and more discriminative assessment of LLM reasoning.
Authors: Huan Song, Deeksha Razdan, Yiyue Qian, Arijit Ghosh Chowdhury, Parth Patwa, Aman Chadha, Shinan Zhang, Sharlina Keshava, Hannah Marlowe
Abstract: Small Language Models (SLMs) offer compelling advantages in deployment cost and latency, but their accuracy often lags behind larger models, particularly for complex domain-specific tasks. While supervised fine-tuning can help bridge this performance gap, it requires substantial manual effort in data preparation and iterative optimization. We present PaDA-Agent (Pattern-guided Data Augmentation Agent), an evaluation-driven approach that streamlines the data augmentation process for SLMs through coordinated operations. Unlike state-of-the-art approaches that focus on model training errors only and generating error-correcting samples, PaDA-Agent discovers failure patterns from the validation data via evaluations and drafts targeted data augmentation strategies aiming to directly reduce the generalization gap. Our experimental results demonstrate significant improvements over state-of-the-art LLM-based data augmentation approaches for Llama 3.2 1B Instruct model fine-tuning.
Authors: Antonio-Gabriel Chac\'on Menke, Phan Xuan Tan, Eiji Kamioka
Abstract: Recent work has highlighted the importance of monitoring chain-of-thought reasoning for AI safety; however, current approaches that analyze textual reasoning steps can miss subtle harmful patterns and may be circumvented by models that hide unsafe reasoning. We present a sentence-level labeled dataset that enables activation-based monitoring of safety behaviors during LLM reasoning. Our dataset contains reasoning sequences with sentence-level annotations of safety behaviors such as expression of safety concerns or speculation on user intent, which we use to extract steering vectors for detecting and influencing these behaviors within model activations. The dataset fills a key gap in safety research: while existing datasets label reasoning holistically, effective application of steering vectors for safety monitoring could be improved by identifying precisely when specific behaviors occur within reasoning chains. We demonstrate the dataset's utility by extracting representations that both detect and steer safety behaviors in model activations, showcasing the potential of activation-level techniques for improving safety oversight on reasoning. Content Warning: This paper discusses AI safety in the context of harmful prompts and may contain references to potentially harmful content.
Authors: Man-Lin Chu, Lucian Terhorst, Kadin Reed, Tom Ni, Weiwei Chen, Rongyu Lin
Abstract: Simulating consumer decision-making is vital for designing and evaluating marketing strategies before costly real- world deployment. However, post-event analyses and rule-based agent-based models (ABMs) struggle to capture the complexity of human behavior and social interaction. We introduce an LLM-powered multi-agent simulation framework that models consumer decisions and social dynamics. Building on recent advances in large language model simulation in a sandbox envi- ronment, our framework enables generative agents to interact, express internal reasoning, form habits, and make purchasing decisions without predefined rules. In a price-discount marketing scenario, the system delivers actionable strategy-testing outcomes and reveals emergent social patterns beyond the reach of con- ventional methods. This approach offers marketers a scalable, low-risk tool for pre-implementation testing, reducing reliance on time-intensive post-event evaluations and lowering the risk of underperforming campaigns.
Authors: Yihong Dong, Zhaoyu Ma, Xue Jiang, Zhiyuan Fan, Jiaru Qian, Yongmin Li, Jianha Xiao, Zhi Jin, Rongyu Cao, Binhua Li, Fei Huang, Yongbin Li, Ge Li
Abstract: Diffusion language models (DLMs) are emerging as a powerful and promising alternative to the dominant autoregressive paradigm, offering inherent advantages in parallel generation and bidirectional context modeling. However, the performance of DLMs on code generation tasks, which have stronger structural constraints, is significantly hampered by the critical trade-off between inference speed and output quality. We observed that accelerating the code generation process by reducing the number of sampling steps usually leads to a catastrophic collapse in performance. In this paper, we introduce efficient Sampling with Adaptive acceleration and Backtracking Enhanced Remasking (i.e., Saber), a novel training-free sampling algorithm for DLMs to achieve better inference speed and output quality in code generation. Specifically, Saber is motivated by two key insights in the DLM generation process: 1) it can be adaptively accelerated as more of the code context is established; 2) it requires a backtracking mechanism to reverse the generated tokens. Extensive experiments on multiple mainstream code generation benchmarks show that Saber boosts Pass@1 accuracy by an average improvement of 1.9% over mainstream DLM sampling methods, meanwhile achieving an average 251.4% inference speedup. By leveraging the inherent advantages of DLMs, our work significantly narrows the performance gap with autoregressive models in code generation.
Authors: Manik Rana, Calissa Man, Anotida Expected Msiiwa, Jeffrey Paine, Kevin Zhu, Sunishchal Dev, Vasu Sharma, Ahan M R
Abstract: Goal changes are a defining feature of real world multi-turn interactions, yet current agent benchmarks primarily evaluate static objectives or one-shot tool use. We introduce AgentChangeBench, a benchmark explicitly designed to measure how tool augmented language model agents adapt to mid dialogue goal shifts across three enterprise domains. Our framework formalizes evaluation through four complementary metrics: Task Success Rate (TSR) for effectiveness, Tool Use Efficiency (TUE) for reliability, Tool Call Redundancy Rate (TCRR) for wasted effort, and Goal-Shift Recovery Time (GSRT) for adaptation latency. AgentChangeBench comprises 2,835 task sequences and five user personas, each designed to trigger realistic shift points in ongoing workflows. Using this setup, we evaluate several frontier models and uncover sharp contrasts obscured by traditional $\text{pass}@k$ scores: for example, GPT-4o reaches $92.2\%$ recovery on airline booking shifts while Gemini collapses to $48.6\%$, and retail tasks show near perfect parameter validity yet redundancy rates above $80\%$, revealing major inefficiencies. These findings demonstrate that high raw accuracy does not imply robustness under dynamic goals, and that explicit measurement of recovery time and redundancy is essential. AgentChangeBench establishes a reproducible testbed for diagnosing and improving agent resilience in realistic enterprise settings.
Authors: Soumya Rani Samineni, Durgesh Kalwar, Vardaan Gangal, Siddhant Bhambri, Subbarao Kambhampati
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR)-based post-training of Large Language Models (LLMs) has been shown to improve accuracy on reasoning tasks and continues to attract significant attention. Existing RLVR methods, however, typically treat all tokens uniformly without accounting for token-level advantages. These methods primarily evaluate performance based on final answer correctness or Pass@K accuracy, and yet make claims about RL post-training leading to improved reasoning traces. This motivates our investigation into the effect of RL post-training on intermediate tokens which are not directly incentivized. To study this, we design an experimental setup using the GRPO algorithm with Qwen-2.5-0.5B model on the GSM8K dataset. We introduce trace coherence, a First-Order Logic (FOL)-based measure to capture the consistency of reasoning steps by identifying errors in the traces. We distinguish between trace validity and trace coherence, noting that the former implies logical soundness while the latter measures local coherence via lack of errors. Our results show that RL post-training overall improves trace coherence with the most significant gains on problems where the base model fails but the RL model succeeds. Surprisingly, RL enhances local coherence without necessarily producing valid or correct solutions. This highlights a crucial distinction: improved local coherence in reasoning steps does not guarantee final answer correctness. We argue that claims of improved reasoning via RL must be examined with care, as these may be based on improved trace coherence, which may not translate into fully valid mathematical proofs.
Authors: Keivan Shariatmadar, Ahmad Osman, Ramin Ray, Usman Dildar, Kisam Kim
Abstract: Fair, transparent, and explainable decision-making remains a critical challenge in Olympic and Paralympic combat sports. This paper presents \emph{FST.ai 2.0}, an explainable AI ecosystem designed to support referees, coaches, and athletes in real time during Taekwondo competitions and training. The system integrates {pose-based action recognition} using graph convolutional networks (GCNs), {epistemic uncertainty modeling} through credal sets, and {explainability overlays} for visual decision support. A set of {interactive dashboards} enables human--AI collaboration in referee evaluation, athlete performance analysis, and Para-Taekwondo classification. Beyond automated scoring, FST.ai~2.0 incorporates modules for referee training, fairness monitoring, and policy-level analytics within the World Taekwondo ecosystem. Experimental validation on competition data demonstrates an {85\% reduction in decision review time} and {93\% referee trust} in AI-assisted decisions. The framework thus establishes a transparent and extensible pipeline for trustworthy, data-driven officiating and athlete assessment. By bridging real-time perception, explainable inference, and governance-aware design, FST.ai~2.0 represents a step toward equitable, accountable, and human-aligned AI in sports.
Authors: Dan Hendrycks, Dawn Song, Christian Szegedy, Honglak Lee, Yarin Gal, Erik Brynjolfsson, Sharon Li, Andy Zou, Lionel Levine, Bo Han, Jie Fu, Ziwei Liu, Jinwoo Shin, Kimin Lee, Mantas Mazeika, Long Phan, George Ingebretsen, Adam Khoja, Cihang Xie, Olawale Salaudeen, Matthias Hein, Kevin Zhao, Alexander Pan, David Duvenaud, Bo Li, Steve Omohundro, Gabriel Alfour, Max Tegmark, Kevin McGrew, Gary Marcus, Jaan Tallinn, Eric Schmidt, Yoshua Bengio
Abstract: The lack of a concrete definition for Artificial General Intelligence (AGI) obscures the gap between today's specialized AI and human-level cognition. This paper introduces a quantifiable framework to address this, defining AGI as matching the cognitive versatility and proficiency of a well-educated adult. To operationalize this, we ground our methodology in Cattell-Horn-Carroll theory, the most empirically validated model of human cognition. The framework dissects general intelligence into ten core cognitive domains-including reasoning, memory, and perception-and adapts established human psychometric batteries to evaluate AI systems. Application of this framework reveals a highly "jagged" cognitive profile in contemporary models. While proficient in knowledge-intensive domains, current AI systems have critical deficits in foundational cognitive machinery, particularly long-term memory storage. The resulting AGI scores (e.g., GPT-4 at 27%, GPT-5 at 58%) concretely quantify both rapid progress and the substantial gap remaining before AGI.
Authors: Xiaohan Qin, Xiaoxing Wang, Ning Liao, Cancheng Zhang, Xiangdong Zhang, Mingquan Feng, Jingzhi Wang, Junchi Yan
Abstract: Data quality plays a critical role in enhancing supervised fine-tuning (SFT) for large language models (LLMs), and token-level data selection has emerged as a promising direction for its fine-grained nature. Despite their strong empirical performance, existing token-level selection methods share two key limitations: (1) requiring training or accessing an additional reference model, and (2) relying solely on loss information for token selection, which cannot well preserve semantically important tokens that are not favored by loss-based metrics. To address these challenges, we propose ssToken, a Self-modulated and Semantic-aware Token Selection approach. ssToken leverages readily accessible history models to compute the per-token loss difference with the current model, which serves as a self-modulated signal that enables the model to adaptively select tokens along its optimization trajectory, rather than relying on excess loss from an offline-trained reference model as in prior works. We further introduce a semantic-aware, attention-based token importance estimation metric, orthogonal to loss-based selection and providing complementary semantic information for more effective filtering. Extensive experiments across different model families and scales demonstrate that both self-modulated selection and semantic-aware selection alone outperform full-data fine-tuning, while their integration--ssToken--achieves synergistic gains and further surpasses prior token-level selection methods, delivering performance improvements while maintaining training efficiency.
Authors: Sion Weatherhead, Flora Salim, Aaron Belbasis
Abstract: Humans do not just find mistakes after the fact -- we often catch them mid-stream because 'reflection' is tied to the goal and its constraints. Today's large language models produce reasoning tokens and 'reflective' text, but is it functionally equivalent with human reflective reasoning? Prior work on closed-ended tasks -- with clear, external 'correctness' signals -- can make 'reflection' look effective while masking limits in self-correction. We therefore test eight frontier models on a simple, real-world task that is open-ended yet rule-constrained, with auditable success criteria: to produce valid scientific test items, then revise after considering their own critique. First-pass performance is poor (often zero valid items out of 4 required; mean $\approx$ 1), and reflection yields only modest gains (also $\approx$ 1). Crucially, the second attempt frequently repeats the same violation of constraint, indicating 'corrective gains' arise largely from chance production of a valid item rather than error detection and principled, constraint-sensitive repair. Performance before and after reflection deteriorates as open-endedness increases, and models marketed for 'reasoning' show no advantage. Our results suggest that current LLM 'reflection' lacks functional evidence of the active, goal-driven monitoring that helps humans respect constraints even on a first pass. Until such mechanisms are instantiated in the model itself, reliable performance requires external structure that enforces constraints.
Authors: Zheng Zhang, Jiarui He, Yuchen Cai, Deheng Ye, Peilin Zhao, Ruili Feng, Hao Wang
Abstract: As large language model (LLM) agents increasingly automate complex web tasks, they boost productivity while simultaneously introducing new security risks. However, relevant studies on web agent attacks remain limited. Existing red-teaming approaches mainly rely on manually crafted attack strategies or static models trained offline. Such methods fail to capture the underlying behavioral patterns of web agents, making it difficult to generalize across diverse environments. In web agent attacks, success requires the continuous discovery and evolution of attack strategies. To this end, we propose Genesis, a novel agentic framework composed of three modules: Attacker, Scorer, and Strategist. The Attacker generates adversarial injections by integrating the genetic algorithm with a hybrid strategy representation. The Scorer evaluates the target web agent's responses to provide feedback. The Strategist dynamically uncovers effective strategies from interaction logs and compiles them into a continuously growing strategy library, which is then re-deployed to enhance the Attacker's effectiveness. Extensive experiments across various web tasks show that our framework discovers novel strategies and consistently outperforms existing attack baselines.
Authors: Aaron Bell, Amit Aides, Amr Helmy, Arbaaz Muslim, Aviad Barzilai, Aviv Slobodkin, Bolous Jaber, David Schottlander, George Leifman, Joydeep Paul, Mimi Sun, Nadav Sherman, Natalie Williams, Per Bjornsson, Roy Lee, Ruth Alcantara, Thomas Turnbull, Tomer Shekel, Vered Silverman, Yotam Gigi, Adam Boulanger, Alex Ottenwess, Ali Ahmadalipour, Anna Carter, Charles Elliott, David Andre, Elad Aharoni, Gia Jung, Hassler Thurston, Jacob Bien, Jamie McPike, Juliet Rothenberg, Kartik Hegde, Kel Markert, Kim Philipp Jablonski, Luc Houriez, Monica Bharel, Phing VanLee, Reuven Sayag, Sebastian Pilarski, Shelley Cazares, Shlomi Pasternak, Siduo Jiang, Stone Jiang, Thomas Colthurst, Yang Chen, Yehonathan Refael, Yochai Blau, Yuval Carny, Yael Maguire, Avinatan Hassidim, James Manyika, Tim Thelin, Genady Beryozkin, Gautam Prasad, Luke Barrington, Yossi Matias, Niv Efron, Shravya Shetty
Abstract: Geospatial data offers immense potential for understanding our planet. However, the sheer volume and diversity of this data along with its varied resolutions, timescales, and sparsity pose significant challenges for thorough analysis and interpretation. This paper introduces Earth AI, a family of geospatial AI models and agentic reasoning that enables significant advances in our ability to unlock novel and profound insights into our planet. This approach is built upon foundation models across three key domains--Planet-scale Imagery, Population, and Environment--and an intelligent Gemini-powered reasoning engine. We present rigorous benchmarks showcasing the power and novel capabilities of our foundation models and validate that when used together, they provide complementary value for geospatial inference and their synergies unlock superior predictive capabilities. To handle complex, multi-step queries, we developed a Gemini-powered agent that jointly reasons over our multiple foundation models along with large geospatial data sources and tools. On a new benchmark of real-world crisis scenarios, our agent demonstrates the ability to deliver critical and timely insights, effectively bridging the gap between raw geospatial data and actionable understanding.
Authors: Peng Tang, Xiaoxiao Yan, Xiaobin Hu, Yuning Cui, Donghao Luo, Jiangning Zhang, Pengcheng Xu, Jinlong Peng, Qingdong He, Feiyue Huang, Song Xue, Tobias Lasser
Abstract: Multi-class unsupervised anomaly detection (MUAD) has garnered growing research interest, as it seeks to develop a unified model for anomaly detection across multiple classes, i.e., eliminating the need to train separate models for distinct objects and thereby saving substantial computational resources. Under the MUAD setting, while advanced Transformer-based architectures have brought significant performance improvements, identity shortcuts persist: they directly copy inputs to outputs, narrowing the gap in reconstruction errors between normal and abnormal cases, and thereby making the two harder to distinguish. Therefore, we propose ShortcutBreaker, a novel unified feature-reconstruction framework for MUAD tasks, featuring two key innovations to address the issue of shortcuts. First, drawing on matrix rank inequality, we design a low-rank noisy bottleneck (LRNB) to project highdimensional features into a low-rank latent space, and theoretically demonstrate its capacity to prevent trivial identity reproduction. Second, leveraging ViTs global modeling capability instead of merely focusing on local features, we incorporate a global perturbation attention to prevent information shortcuts in the decoders. Extensive experiments are performed on four widely used anomaly detection benchmarks, including three industrial datasets (MVTec-AD, ViSA, and Real-IAD) and one medical dataset (Universal Medical). The proposed method achieves a remarkable image-level AUROC of 99.8%, 98.9%, 90.6%, and 87.8% on these four datasets, respectively, consistently outperforming previous MUAD methods across different scenarios.
Authors: Runnan Qi, Yanan Ni, Lumin Jiang, Zongyuan Li, Kuihua Huang, Xian Guo
Abstract: This paper proposes Memory-Augmented State Machine Prompting (MASMP), a novel framework for LLM agents in real-time strategy games. Addressing key challenges like hallucinations and fragmented decision-making in existing approaches, MASMP integrates state machine prompting with memory mechanisms to unify structured actions with long-term tactical coherence. The framework features: (1) a natural language-driven state machine architecture that guides LLMs to emulate finite state machines and behavior trees through prompts, and (2) a lightweight memory module preserving strategic variables (e.g., tactics, priority units) across decision cycles. Experiments in StarCraft II demonstrate MASMP's 60% win rate against the hardest built-in AI (Lv7), vastly outperforming baselines (0%). Case studies reveal the method retains LLMs' semantic comprehension while resolving the "Knowing-Doing Gap" through strict state-action mapping, achieving both interpretability and FSM-like reliability. This work establishes a new paradigm for combining neural and symbolic AI in complex decision-making.
Authors: Manjie Xu, Xinyi Yang, Jiayu Zhan, Wei Liang, Chi Zhang, Yixin Zhu
Abstract: Self-play constitutes a fundamental paradigm for autonomous skill acquisition, whereby agents iteratively enhance their capabilities through self-directed environmental exploration. Conventional self-play frameworks exploit agent symmetry within zero-sum competitive settings, yet this approach proves inadequate for open-ended learning scenarios characterized by inherent asymmetry. Human pedagogical systems exemplify asymmetric instructional frameworks wherein educators systematically construct challenges calibrated to individual learners' developmental trajectories. The principal challenge resides in operationalizing these asymmetric, adaptive pedagogical mechanisms within artificial systems capable of autonomously synthesizing appropriate curricula without predetermined task hierarchies. Here we present Heterogeneous Adversarial Play (HAP), an adversarial Automatic Curriculum Learning framework that formalizes teacher-student interactions as a minimax optimization wherein task-generating instructor and problem-solving learner co-evolve through adversarial dynamics. In contrast to prevailing ACL methodologies that employ static curricula or unidirectional task selection mechanisms, HAP establishes a bidirectional feedback system wherein instructors continuously recalibrate task complexity in response to real-time learner performance metrics. Experimental validation across multi-task learning domains demonstrates that our framework achieves performance parity with SOTA baselines while generating curricula that enhance learning efficacy in both artificial agents and human subjects.
Authors: Mattia Pujatti, Andrea Di Luca, Nicola Peghini, Federico Monegaglia, Marco Cristoforetti
Abstract: In glass bottle manufacturing, precise control of forming machines is critical for ensuring quality and minimizing defects. This study presents a deep learning-based control algorithm designed to optimize the forming process in real production environments. Using real operational data from active manufacturing plants, our neural network predicts the effects of parameter changes based on the current production setup. Through a specifically designed inversion mechanism, the algorithm identifies the optimal machine settings required to achieve the desired glass gob characteristics. Experimental results on historical datasets from multiple production lines show that the proposed method yields promising outcomes, suggesting potential for enhanced process stability, reduced waste, and improved product consistency. These results highlight the potential of deep learning to process control in glass manufacturing.
Authors: Guangfu Guo, Xiaoqian Lu, Yue Feng
Abstract: Visual Language Models (VLMs) achieve promising results in medical reasoning but struggle with hallucinations, vague descriptions, inconsistent logic and poor localization. To address this, we propose a agent framework named Medical Visual Reasoning Agent (\textbf{Med-VRAgent}). The approach is based on Visual Guidance and Self-Reward paradigms and Monte Carlo Tree Search (MCTS). By combining the Visual Guidance with tree search, Med-VRAgent improves the medical visual reasoning capabilities of VLMs. We use the trajectories collected by Med-VRAgent as feedback to further improve the performance by fine-tuning the VLMs with the proximal policy optimization (PPO) objective. Experiments on multiple medical VQA benchmarks demonstrate that our method outperforms existing approaches.
Authors: Chenxu Zhang, Fuxiang Huang, Lei Zhang
Abstract: With climate change intensifying, urban waterlogging poses an increasingly severe threat to global public safety and infrastructure. However, existing monitoring approaches rely heavily on manual reporting and fail to provide timely and comprehensive assessments. In this study, we present Urban Waterlogging Assessment (UWAssess), a foundation model-driven framework that automatically identifies waterlogged areas in surveillance images and generates structured assessment reports. To address the scarcity of labeled data, we design a semi-supervised fine-tuning strategy and a chain-of-thought (CoT) prompting strategy to unleash the potential of the foundation model for data-scarce downstream tasks. Evaluations on challenging visual benchmarks demonstrate substantial improvements in perception performance. GPT-based evaluations confirm the ability of UWAssess to generate reliable textual reports that accurately describe waterlogging extent, depth, risk and impact. This dual capability enables a shift of waterlogging monitoring from perception to generation, while the collaborative framework of multiple foundation models lays the groundwork for intelligent and scalable systems, supporting urban management, disaster response and climate resilience.
Authors: Minwei Kong, Ao Qu, Xiaotong Guo, Wenbin Ouyang, Chonghe Jiang, Han Zheng, Yining Ma, Dingyi Zhuang, Yuhan Tang, Junyi Li, Hai Wang, Cathy Wu, Jinhua Zhao
Abstract: Optimization modeling enables critical decisions across industries but remains difficult to automate: informal language must be mapped to precise mathematical formulations and executable solver code. Prior LLM approaches either rely on brittle prompting or costly retraining with limited generalization. We present AlphaOPT, a self-improving experience library that enables an LLM to learn from limited demonstrations (even answers alone, without gold-standard programs) and solver feedback - without annotated reasoning traces or parameter updates. AlphaOPT operates in a continual two-phase cycle: (i) a Library Learning phase that reflects on failed attempts, extracting solver-verified, structured insights as {taxonomy, condition, explanation, example}; and (ii) a Library Evolution phase that diagnoses retrieval misalignments and refines the applicability conditions of stored insights, improving transfer across tasks. This design (1) learns efficiently from limited demonstrations without curated rationales, (2) expands continually without costly retraining by updating the library rather than model weights, and (3) makes knowledge explicit and interpretable for human inspection and intervention. Experiments show that AlphaOPT steadily improves with more data (65% to 72% from 100 to 300 training items) and surpasses the strongest baseline by 7.7% on the out-of-distribution OptiBench dataset when trained only on answers. Code and data are available at: https://github.com/Minw913/AlphaOPT.
Authors: Ziwei Deng, Mian Deng, Chenjing Liang, Zeming Gao, Chennan Ma, Chenxing Lin, Haipeng Zhang, Songzhu Mei, Cheng Wang, Siqi Shen
Abstract: Large Language Models (LLMs) are increasingly being explored across a range of decision-making tasks. However, LLMs sometimes struggle with decision-making tasks under uncertainty that are relatively easy for humans, such as planning actions in stochastic environments. The adoption of LLMs for decision-making is impeded by uncertainty challenges, such as LLM uncertainty and environmental uncertainty. LLM uncertainty arises from the stochastic sampling process inherent to LLMs. Most LLM-based Decision-Making (LDM) approaches address LLM uncertainty through multiple reasoning chains or search trees. However, these approaches overlook environmental uncertainty, which leads to poor performance in environments with stochastic state transitions. Some recent LDM approaches deal with uncertainty by forecasting the probability of unknown variables. However, they are not designed for multi-step decision-making tasks that require interaction with the environment. To address uncertainty in LLM decision-making, we introduce PlanU, an LLM-based planning method that captures uncertainty within Monte Carlo Tree Search (MCTS). PlanU models the return of each node in the MCTS as a quantile distribution, which uses a set of quantiles to represent the return distribution. To balance exploration and exploitation during tree search, PlanU introduces an Upper Confidence Bounds with Curiosity (UCC) score which estimates the uncertainty of MCTS nodes. Through extensive experiments, we demonstrate the effectiveness of PlanU in LLM-based decision-making tasks under uncertainty.
Authors: Shaobo Wang, Yongliang Miao, Yuancheng Liu, and Qianli Ma, Ning Liao, Linfeng Zhang
Abstract: Large language models (LLMs) have demonstrated impressive reasoning capabilities, but scaling their performance often relies on massive reasoning datasets that are computationally expensive to train on. Existing data selection methods aim to curate smaller, high-quality subsets but often rely on costly external models or opaque heuristics. In this work, we shift the focus from external heuristics to the model's internal mechanisms. We find that complex reasoning tasks consistently activate a sparse, specialized subset of attention heads, forming core reasoning circuits. Building on this insight, we propose CircuitSeer, a novel data selection method that quantifies the reasoning complexity of data by measuring its influence on these crucial circuits. Extensive experiments on 4 models and 9 datasets demonstrate CircuitSeer's superiority. Notably, fine-tuning Qwen2.5-Math-7B on just 10% of data selected by our method achieves a 1.4-point gain in average Pass@1 over training on the full dataset, highlighting its efficiency and effectiveness.
Authors: Feifan Xia, Yuyang Fang, Defang Li, Yantong Xie, Weikang Li, Yang Li, Deguo Xia, Jizhou Huang
Abstract: We present a probabilistic intent modeling framework for large language model (LLM) agents in multi-turn social dialogue. The framework maintains a belief distribution over a partner's latent intentions, initialized from contextual priors and dynamically updated through likelihood estimation after each utterance. The evolving distribution provides additional contextual grounding for the policy, enabling adaptive dialogue strategies under uncertainty. Preliminary experiments in the SOTOPIA environment show consistent improvements: the proposed framework increases the Overall score by 9.0% on SOTOPIA-All and 4.1% on SOTOPIA-Hard compared with the Qwen2.5-7B baseline, and slightly surpasses an oracle agent that directly observes partner intentions. These early results suggest that probabilistic intent modeling can contribute to the development of socially intelligent LLM agents.
Authors: Haichao Ji, Zibo Wang, Yifei Zhu, Meng han, Dan Wang, Zhu Han
Abstract: Large Language Models (LLMs) have shown great promise in automating data analytics tasks by interpreting natural language queries and generating multi-operation execution plans. However, existing LLM-agent-based analytics frameworks operate under the assumption of centralized data access, offering little to no privacy protection. In contrast, federated analytics (FA) enables privacy-preserving computation across distributed data sources, but lacks support for natural language input and requires structured, machine-readable queries. In this work, we present LAFA, the first system that integrates LLM-agent-based data analytics with FA. LAFA introduces a hierarchical multi-agent architecture that accepts natural language queries and transforms them into optimized, executable FA workflows. A coarse-grained planner first decomposes complex queries into sub-queries, while a fine-grained planner maps each subquery into a Directed Acyclic Graph of FA operations using prior structural knowledge. To improve execution efficiency, an optimizer agent rewrites and merges multiple DAGs, eliminating redundant operations and minimizing computational and communicational overhead. Our experiments demonstrate that LAFA consistently outperforms baseline prompting strategies by achieving higher execution plan success rates and reducing resource-intensive FA operations by a substantial margin. This work establishes a practical foundation for privacy-preserving, LLM-driven analytics that supports natural language input in the FA setting.
Authors: Haoran Zhang, Chenhao Zhu, Sicong Guo, Hanzhe Guo, Haiming Li, Donglin Yu
Abstract: Human players do more than press buttons: they ground what they see on screen into precise keyboard-mouse actions and, when stuck, they seek information before trying again. We ask whether current vision-language models (VLMs) can do the same. Despite encouraging results under simplified control or tool scaffolds, human-like play in a real client - mapping raw screenshots to temporally coherent low-level actions while deciding when to ask for guidance - remains an open challenge. We introduce StarBench, a turn-based RPG benchmark derived from Honkai: Star Rail that targets these two human-like competencies: multimodal decision-making from pixels to actions and agentic information seeking. StarBench standardizes evaluation across eight combat tasks and two regimes with shared tasks and metrics: (i) direct control, where agents receive only screenshots and must emit low-level primitives (click and keypress) with no semantic hints; and (ii) tool-assisted control, where higher-level intents can be mapped to primitives by detectors and OCR outputs provide optional textualized observations to ease UI grounding. To mirror human practice, StarBench also includes an ask-or-act diagnostic that measures whether and when agents choose to request brief guidance before proceeding, and how that choice affects subsequent performance. We report reference baselines for contemporary VLMs and a human reference. Results expose sizable gaps in perception-to-control fidelity in the direct regime, while showing that judicious information seeking correlates with improved success, establishing StarBench as a reproducible yardstick for agentic information seeking and multimodal decision-making in real-client play.
Authors: Ho Fai Leung, Xiaoyan Xi, Fei Zuo
Abstract: On-device virtual assistants like Siri and Google Assistant are increasingly pivotal, yet their capabilities are hamstrung by a reliance on rigid, developer-dependent APIs. GUI agents offer a powerful, API-independent alternative, but their adoption is hindered by the perception of poor performance, as even the best models (e.g. Qwen3-VL-235B) scores are capped at around 60% on benchmarks like AndroidControl, far from viability for real-world use. Our research reveals that issue lies not only with the models but with the benchmarks themselves. We identified notable shortcomings in AndroidControl, including ambiguities and factual errors, which systematically underrates agent capabilities. To address this critical oversight, we enhanced AndroidControl into AndroidControl-Curated, a refined version of the benchmark improved through a rigorous purification pipeline. On this enhanced benchmark, state-of-the-art models achieve success rates nearing 75% on complex tasks (15% improvement), reflecting that on-device GUI agents are actually closer to practical deployment than previously thought. We introduce our new SOTA model, Magma-R1- 3B, post-trained on just 2.4k curated samples using 60 hours of an H20 GPU (approximately $60). Despite being 200 times smaller in parameters, this model delivers performance comparable to Qwen3- VL-235B. We release both AndroidControl-Curated benchmark and Magma-R1 model to the research community, encouraging adoption of this enhanced benchmark to better reflect model capabilities and accelerate the development of robust, on-device virtual assistants.
Authors: Lianchen Jia, Chaoyang Li, Qian Houde, Tianchi Huang, Jiangchuan Liu, Lifeng Sun
Abstract: Control algorithms in production environments typically require domain experts to tune their parameters and logic for specific scenarios. However, existing research predominantly focuses on algorithmic performance under ideal or default configurations, overlooking the critical aspect of Tuning Potential. To bridge this gap, we introduce Crucible, an agent that employs an LLM-driven, multi-level expert simulation to turn algorithms and defines a formalized metric to quantitatively evaluate their Tuning Potential. We demonstrate Crucible's effectiveness across a wide spectrum of case studies, from classic control tasks to complex computer systems, and validate its findings in a real-world deployment. Our experimental results reveal that Crucible systematically quantifies the tunable space across different algorithms. Furthermore, Crucible provides a new dimension for algorithm analysis and design, which ultimately leads to performance improvements. Our code is available at https://github.com/thu-media/Crucible.
Authors: Hanze Guo, Jing Yao, Xiao Zhou, Xiaoyuan Yi, Xing Xie
Abstract: As large language models (LLMs) become increasingly integrated into applications serving users across diverse cultures, communities and demographics, it is critical to align LLMs with pluralistic human values beyond average principles (e.g., HHH). In psychological and social value theories such as Schwartz's Value Theory, pluralistic values are represented by multiple value dimensions paired with various priorities. However, existing methods encounter two challenges when aligning with such fine-grained value objectives: 1) they often treat multiple values as independent and equally important, ignoring their interdependence and relative priorities (value complexity); 2) they struggle to precisely control nuanced value priorities, especially those underrepresented ones (value steerability). To handle these challenges, we propose COUPLE, a COUnterfactual reasoning framework for PLuralistic valuE alignment. It introduces a structural causal model (SCM) to feature complex interdependency and prioritization among features, as well as the causal relationship between high-level value dimensions and behaviors. Moreover, it applies counterfactual reasoning to generate outputs aligned with any desired value objectives. Benefitting from explicit causal modeling, COUPLE also provides better interpretability. We evaluate COUPLE on two datasets with different value systems and demonstrate that COUPLE advances other baselines across diverse types of value objectives.
Authors: Sarth Dubey, Subimal Ghosh, Udit Bhatia
Abstract: Reliable hydrologic and flood forecasting requires models that remain stable when input data are delayed, missing, or inconsistent. However, most advances in rainfall-runoff prediction have been evaluated under ideal data conditions, emphasizing accuracy rather than operational resilience. Here, we develop an operationally ready emulator of the Global Flood Awareness System (GloFAS) that couples long- and short-term memory networks with a relaxed water-balance constraint to preserve physical coherence. Five architectures span a continuum of information availability: from complete historical and forecast forcings to scenarios with data latency and outages, allowing systematic evaluation of robustness. Trained in minimally managed catchments across the United States and tested in more than 5,000 basins, including heavily regulated rivers in India, the emulator reproduces the hydrological core of GloFAS and degrades smoothly as information quality declines. Transfer across contrasting hydroclimatic and management regimes yields reduced yet physically consistent performance, defining the limits of generalization under data scarcity and human influence. The framework establishes operational robustness as a measurable property of hydrological machine learning and advances the design of reliable real-time forecasting systems.
Authors: Yuncheng Hua, Sion Weatherhead, Mehdi Jafari, Hao Xue, Flora D. Salim
Abstract: In this paper, we present SOCIA-Nabla, an end-to-end, agentic framework that treats simulator construction asinstance optimization over code within a textual computation graph. Specialized LLM-driven agents are embedded as graph nodes, and a workflow manager executes a loss-driven loop: code synthesis -> execution -> evaluation -> code repair. The optimizer performs Textual-Gradient Descent (TGD), while human-in-the-loop interaction is reserved for task-spec confirmation, minimizing expert effort and keeping the code itself as the trainable object. Across three CPS tasks, i.e., User Modeling, Mask Adoption, and Personal Mobility, SOCIA-Nabla attains state-of-the-art overall accuracy. By unifying multi-agent orchestration with a loss-aligned optimization view, SOCIA-Nabla converts brittle prompt pipelines into reproducible, constraint-aware simulator code generation that scales across domains and simulation granularities. This work is under review, and we will release the code soon.
Authors: Federico Barbero, Xiangming Gu, Christopher A. Choquette-Choo, Chawin Sitawarin, Matthew Jagielski, Itay Yona, Petar Veli\v{c}kovi\'c, Ilia Shumailov, Jamie Hayes
Abstract: In this work, we show that it is possible to extract significant amounts of alignment training data from a post-trained model -- useful to steer the model to improve certain capabilities such as long-context reasoning, safety, instruction following, and maths. While the majority of related work on memorisation has focused on measuring success of training data extraction through string matching, we argue that embedding models are better suited for our specific goals. Distances measured through a high quality embedding model can identify semantic similarities between strings that a different metric such as edit distance will struggle to capture. In fact, in our investigation, approximate string matching would have severely undercounted (by a conservative estimate of $10\times$) the amount of data that can be extracted due to trivial artifacts that deflate the metric. Interestingly, we find that models readily regurgitate training data that was used in post-training phases such as SFT or RL. We show that this data can be then used to train a base model, recovering a meaningful amount of the original performance. We believe our work exposes a possibly overlooked risk towards extracting alignment data. Finally, our work opens up an interesting discussion on the downstream effects of distillation practices: since models seem to be regurgitating aspects of their training set, distillation can therefore be thought of as indirectly training on the model's original dataset.
Authors: Junhyeog Yun, Hyoun Jun Lee, Insu Jeon
Abstract: Automating quantitative trading strategy development in dynamic markets is challenging, especially with increasing demand for personalized investment solutions. Existing methods often fail to explore the vast strategy space while preserving the diversity essential for robust performance across changing market conditions. We present QuantEvolve, an evolutionary framework that combines quality-diversity optimization with hypothesis-driven strategy generation. QuantEvolve employs a feature map aligned with investor preferences, such as strategy type, risk profile, turnover, and return characteristics, to maintain a diverse set of effective strategies. It also integrates a hypothesis-driven multi-agent system to systematically explore the strategy space through iterative generation and evaluation. This approach produces diverse, sophisticated strategies that adapt to both market regime shifts and individual investment needs. Empirical results show that QuantEvolve outperforms conventional baselines, validating its effectiveness. We release a dataset of evolved strategies to support future research.
Authors: Wei Cai, Jian Zhao, Yuchen Yuan, Tianle Zhang, Ming Zhu, Haichuan Tang, Chi Zhang, Xuelong Li
Abstract: Multimodal Large Language Models (MLLMs), despite their advances, are hindered by their high hallucination tendency and heavy reliance on brittle, linear reasoning processes, leading to failures in complex tasks. To address these limitations, we introduce Visual Attention Reasoning (VAR), a novel framework that recasts grounded reasoning as a structured search over a reasoning trajectory space. VAR decomposes the reasoning process into two key stages: traceable evidence grounding and search-based chain-of-thought (CoT) generation, which incorporates a backtracking mechanism for self-correction. The search is guided by a multi-faceted reward function with semantic and geometric self-verification components, which penalize outputs that are not faithfully grounded in the visual input. We provide a theoretical analysis for our search strategy, validating its capability to find the correct solution with high probability. Experimental results show that our 7B model, VAR-7B, sets a new state-of-the-art on a comprehensive suite of hallucination and safety benchmarks, significantly outperforming existing open-source models and demonstrating competitive performance against leading proprietary systems.
Authors: Gilles Audemard, Sylvie Coste-Marquis, Pierre Marquis, Mehdi Sabiri, Nicolas Szczepanski
Abstract: We present a new approach to classification that combines data and knowledge. In this approach, data mining is used to derive association rules (possibly with negations) from data. Those rules are leveraged to increase the predictive performance of tree-based models (decision trees and random forests) used for a classification task. They are also used to improve the corresponding explanation task through the generation of abductive explanations that are more general than those derivable without taking such rules into account. Experiments show that for the two tree-based models under consideration, benefits can be offered by the approach in terms of predictive performance and in terms of explanation sizes.
Authors: Carlo Proietti, Antonio Yuste-Ginel
Abstract: Modelling qualitative uncertainty in formal argumentation is essential both for practical applications and theoretical understanding. Yet, most of the existing works focus on \textit{abstract} models for arguing with uncertainty. Following a recent trend in the literature, we tackle the open question of studying plausible instantiations of these abstract models. To do so, we ground the uncertainty of arguments in their components, structured within rules and premises. Our main technical contributions are: i) the introduction of a notion of expressivity that can handle abstract and structured formalisms, and ii) the presentation of both negative and positive expressivity results, comparing the expressivity of abstract and structured models of argumentation with uncertainty. These results affect incomplete abstract argumentation frameworks, and their extension with dependencies, on the abstract side, and ASPIC+, on the structured side.
Authors: Roxana Petcu, Kenton Murray, Daniel Khashabi, Evangelos Kanoulas, Maarten de Rijke, Dawn Lawrie, Kevin Duh
Abstract: Retrieval-augmented generation (RAG) systems address complex user requests by decomposing them into subqueries, retrieving potentially relevant documents for each, and then aggregating them to generate an answer. Efficiently selecting informative documents requires balancing a key trade-off: (i) retrieving broadly enough to capture all the relevant material, and (ii) limiting retrieval to avoid excessive noise and computational cost. We formulate query decomposition and document retrieval in an exploitation-exploration setting, where retrieving one document at a time builds a belief about the utility of a given sub-query and informs the decision to continue exploiting or exploring an alternative. We experiment with a variety of bandit learning methods and demonstrate their effectiveness in dynamically selecting the most informative sub-queries. Our main finding is that estimating document relevance using rank information and human judgments yields a 35% gain in document-level precision, 15% increase in {\alpha}-nDCG, and better performance on the downstream task of long-form generation.
Authors: Dong Yun, Marco Schouten, Dim Papadopoulos
Abstract: User queries in information retrieval are often ambiguous, making it challenging for systems to identify a user's target from a single query. While recent dialogue-based interactive retrieval systems can clarify user intent, they are inefficient as they often lack an explicit strategy to ask the most informative questions. To address this limitation, we propose SherlockLLM, a dialogue-driven retrieval framework that learns an optimal questioning strategy via Reinforcement Learning (RL) and avoids the need for large-scale annotated dialogue data. In our framework, an agent is trained to generate a sequence of binary questions to efficiently narrow down the search space. To validate our approach, we introduce a benchmark with both structured and unstructured tasks. Experimental results show that SherlockLLM is a robust and efficient solution. On the structured tasks, its performance matches strong baselines and approaches the theoretical optimal defined by binary search. On the challenging unstructured task, our agent significantly outperforms these baselines, showcasing its ability to learn a highly effective information-seeking dialogue policy.
Authors: Patterson Hsieh, Jerry Yeh, Mao-Chi He, Wen-Han Hsieh, Elvis Hsieh
Abstract: Climate change is intensifying the occurrence of harmful algal bloom (HAB), particularly cyanobacteria, which threaten aquatic ecosystems and human health through oxygen depletion, toxin release, and disruption of marine biodiversity. Traditional monitoring approaches, such as manual water sampling, remain labor-intensive and limited in spatial and temporal coverage. Recent advances in vision-language models (VLMs) for remote sensing have shown potential for scalable AI-driven solutions, yet challenges remain in reasoning over imagery and quantifying bloom severity. In this work, we introduce ALGae Observation and Segmentation (ALGOS), a segmentation-and-reasoning system for HAB monitoring that combines remote sensing image understanding with severity estimation. Our approach integrates GeoSAM-assisted human evaluation for high-quality segmentation mask curation and fine-tunes vision language model on severity prediction using the Cyanobacteria Aggregated Manual Labels (CAML) from NASA. Experiments demonstrate that ALGOS achieves robust performance on both segmentation and severity-level estimation, paving the way toward practical and automated cyanobacterial monitoring systems.
Authors: Shirin Tavakoli Kafiabad, Andrea Schiffauerova, Ashkan Ebadi
Abstract: Optimizing national scientific investment requires a clear understanding of evolving research trends and the demographic and geographical forces shaping them, particularly in light of commitments to equity, diversity, and inclusion. This study addresses this need by analyzing 18 years (2005-2022) of research proposals funded by the Natural Sciences and Engineering Research Council of Canada (NSERC). We conducted a comprehensive comparative evaluation of three topic modelling approaches: Latent Dirichlet Allocation (LDA), Structural Topic Modelling (STM), and BERTopic. We also introduced a novel algorithm, named COFFEE, designed to enable robust covariate effect estimation for BERTopic. This advancement addresses a significant gap, as BERTopic lacks a native function for covariate analysis, unlike the probabilistic STM. Our findings highlight that while all models effectively delineate core scientific domains, BERTopic outperformed by consistently identifying more granular, coherent, and emergent themes, such as the rapid expansion of artificial intelligence. Additionally, the covariate analysis, powered by COFFEE, confirmed distinct provincial research specializations and revealed consistent gender-based thematic patterns across various scientific disciplines. These insights offer a robust empirical foundation for funding organizations to formulate more equitable and impactful funding strategies, thereby enhancing the effectiveness of the scientific ecosystem.
Authors: Xinsheng Wang, Shanmin Pang, Jihua Zhu, Zhongyu Li, Zhiqiang Tian, Yaochen Li
Abstract: Zero-shot learning, which aims to recognize new categories that are not included in the training set, has gained popularity owing to its potential ability in the real-word applications. Zero-shot learning models rely on learning an embedding space, where both semantic descriptions of classes and visual features of instances can be embedded for nearest neighbor search. Recently, most of the existing works consider the visual space formulated by deep visual features as an ideal choice of the embedding space. However, the discrete distribution of instances in the visual space makes the data structure unremarkable. We argue that optimizing the visual space is crucial as it allows semantic vectors to be embedded into the visual space more effectively. In this work, we propose two strategies to accomplish this purpose. One is the visual prototype based method, which learns a visual prototype for each visual class, so that, in the visual space, a class can be represented by a prototype feature instead of a series of discrete visual features. The other is to optimize the visual feature structure in an intermediate embedding space, and in this method we successfully devise a multilayer perceptron framework based algorithm that is able to learn the common intermediate embedding space and meanwhile to make the visual data structure more distinctive. Through extensive experimental evaluation on four benchmark datasets, we demonstrate that optimizing visual space is beneficial for zero-shot learning. Besides, the proposed prototype based method achieves the new state-of-the-art performance.
Authors: Qun Wang, Yingzhou Lu, Guiran Liu, Binrong Zhu, Yang Liu
Abstract: Unlicensed 6GHz is becoming a primary workhorse for high-capacity access, with Wi-Fi and 5G NR-U competing for the same channels under listen-before-talk (LBT) rules. Operating in this regime requires decisions that jointly trade throughput, energy, and service-level objectives while remaining safe and auditable. We present an agentic controller that separates {policy} from {execution}. At the start of each scheduling epoch the agent summarizes telemetry (per-channel busy and baseline LBT failure; per-user CQI, backlog, latency, battery, priority, and power mode) and invokes a large language model (LLM) to propose a small set of interpretable knobs: a fairness index \alpha, per-channel duty-cycle caps for Wi-Fi/NR-U, and class weights. A deterministic optimizer then enforces feasibility and computes an \alpha-fair allocation that internalizes LBT losses and energy cost; malformed or unsafe policies are clamped and fall back to a rule baseline. In a 6GHz simulator with two 160MHz channels and mixed Wi-Fi/NR-U users, LLM-assisted policies consistently improve energy efficiency while keeping throughput competitive with a strong rule baseline. One LLM lowers total energy by 35.3% at modest throughput loss, and another attains the best overall trade-off, finishing with higher total bits (+3.5%) and higher bits/J (+12.2%) than the baseline. We release code, per-epoch logs, and plotting utilities to reproduce all figures and numbers, illustrating how transparent, policy-level LLM guidance can safely improve wireless coexistence.
Authors: D. Halatsis, P. Mamidanna, J. Pereira, D. Farina
Abstract: Recent advances in neural interfacing have enabled significant improvements in human-computer interaction, rehabilitation, and neuromuscular diagnostics. Motor unit (MU) decomposition from surface electromyography (sEMG) is a key technique for extracting neural drive information, but traditional blind source separation (BSS) methods fail to incorporate biophysical constraints, limiting their accuracy and interpretability. In this work, we introduce a novel Biophysical-Model-Informed Source Separation (BMISS) framework, which integrates anatomically accurate forward EMG models into the decomposition process. By leveraging MRI-based anatomical reconstructions and generative modeling, our approach enables direct inversion of a biophysically accurate forward model to estimate both neural drive and motor neuron properties in an unsupervised manner. Empirical validation in a controlled simulated setting demonstrates that BMISS achieves higher fidelity motor unit estimation while significantly reducing computational cost compared to traditional methods. This framework paves the way for non-invasive, personalized neuromuscular assessments, with potential applications in clinical diagnostics, prosthetic control, and neurorehabilitation.
Authors: Shumaila Javaid, Nasir Saeed
Abstract: Integrated Satellite Aerial Terrestrial Networks (ISATNs) are envisioned as key enablers of 6G, providing global connectivity for applications such as autonomous transportation, Industrial IoT, and disaster response. Their large-scale deployment, however, risks unsustainable energy use and carbon emissions. This work advances prior energy-aware studies by proposing a carbon-aware orchestration framework for ISATNs that leverages Digital Twin (DT) technology. The framework adopts grams of CO$_2$-equivalent per bit (gCO$_2$/bit) as a primary sustainability metric and implements a multi timescale Plan Do Check Act (PDCA) loop that combines day-ahead forecasting with real-time adaptive optimization. ISATN-specific control knobs, including carbon-aware handovers, UAV duty cycling, and renewable-aware edge placement, are exploited to reduce emissions. Simulation results with real carbon intensity data show up to 29\% lower gCO$_2$/bit than QoS-only orchestration, while improving renewable utilization and resilience under adverse events.
Authors: Carles Navarro, Mariona Torrens, Philipp Th\"olke, Stefan Doerr, Gianni De Fabritiis
Abstract: Building a working mental model of a protein typically requires weeks of reading, cross-referencing crystal and predicted structures, and inspecting ligand complexes, an effort that is slow, unevenly accessible, and often requires specialized computational skills. We introduce \emph{Speak to a Protein}, a new capability that turns protein analysis into an interactive, multimodal dialogue with an expert co-scientist. The AI system retrieves and synthesizes relevant literature, structures, and ligand data; grounds answers in a live 3D scene; and can highlight, annotate, manipulate and see the visualization. It also generates and runs code when needed, explaining results in both text and graphics. We demonstrate these capabilities on relevant proteins, posing questions about binding pockets, conformational changes, or structure-activity relationships to test ideas in real-time. \emph{Speak to a Protein} reduces the time from question to evidence, lowers the barrier to advanced structural analysis, and enables hypothesis generation by tightly coupling language, code, and 3D structures. \emph{Speak to a Protein} is freely accessible at https://open.playmolecule.org.
Authors: Meir H. Shachar, Dane M. Sterbentz, Harshitha Menon, Charles F. Jekel, M. Giselle Fern\'andez-Godino, Yue Hao, Kevin Korner, Robert Rieben, Daniel A. White, William J. Schill, Jonathan L. Belof
Abstract: Inertial fusion energy promises nearly unlimited, clean power if it can be achieved. However, the design and engineering of fusion systems requires controlling and manipulating matter at extreme energies and timescales; the shock physics and radiation transport governing the physical behavior under these conditions are complex requiring the development, calibration, and use of predictive multiphysics codes to navigate the highly nonlinear and multi-faceted design landscape. We hypothesize that artificial intelligence reasoning models can be combined with physics codes and emulators to autonomously design fusion fuel capsules. In this article, we construct a multi-agent system where natural language is utilized to explore the complex physics regimes around fusion energy. The agentic system is capable of executing a high-order multiphysics inertial fusion computational code. We demonstrate the capacity of the multi-agent design assistant to both collaboratively and autonomously manipulate, navigate, and optimize capsule geometry while accounting for high fidelity physics that ultimately achieve simulated ignition via inverse design.
Authors: Henrique de Lima Alexandre, Clodoaldo Aparecido de Moraes Lima
Abstract: Electroencephalography (EEG) is a widely used, non-invasive method for capturing brain activity, and is particularly relevant for applications in Brain-Computer Interfaces (BCI). However, collecting high-quality EEG data remains a major challenge due to sensor costs, acquisition time, and inter-subject variability. To address these limitations, this study proposes a methodology for generating synthetic EEG signals associated with motor imagery brain tasks using Diffusion Probabilistic Models (DDPM). The approach involves preprocessing real EEG data, training a diffusion model to reconstruct EEG channels from noise, and evaluating the quality of the generated signals through both signal-level and task-level metrics. For validation, we employed classifiers such as K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNN), and U-Net to compare the performance of synthetic data against real data in classification tasks. The generated data achieved classification accuracies above 95%, with low mean squared error and high correlation with real signals. Our results demonstrate that synthetic EEG signals produced by diffusion models can effectively complement datasets, improving classification performance in EEG-based BCIs and addressing data scarcity.
Authors: \'Angela L\'opez-Cardona, Sebasti\'an Idesis, Mireia Masias-Bruns, Sergi Abadal, Ioannis Arapakis
Abstract: Do brains and language models converge toward the same internal representations of the world? Recent years have seen a rise in studies of neural activations and model alignment. In this work, we review 25 fMRI-based studies published between 2023 and 2025 and explicitly confront their findings with two key hypotheses: (i) the Platonic Representation Hypothesis -- that as models scale and improve, they converge to a representation of the real world, and (ii) the Intermediate-Layer Advantage -- that intermediate (mid-depth) layers often encode richer, more generalizable features. Our findings provide converging evidence that models and brains may share abstract representational structures, supporting both hypotheses and motivating further research on brain-model alignment.
Authors: Zongze Wu, Yani Guo, Churong Liang, Runnan Li
Abstract: Despite remarkable advances in Large Language Model capabilities, tool retrieval for agent-based systems remains fundamentally limited by reliance on semantic similarity, which fails to capture functional viability. Current methods often retrieve textually relevant but functionally inoperative tools due to parameter mismatches, authentication failures, and execution constraints--a phenomenon we term the semantic-functional gap. We introduce GRETEL, to address this gap through systematic empirical validation. GRETEL implements an agentic workflow that processes semantically retrieved candidates through sandboxed plan-execute-evaluate cycles, generating execution-grounded evidence to distinguish truly functional tools from merely descriptive matches. Our comprehensive evaluation on the ToolBench benchmark demonstrates substantial improvements across all metrics: Pass Rate (at 10) increases from 0.690 to 0.826, Recall (at 10) improves from 0.841 to 0.867, and NDCG (at 10) rises from 0.807 to 0.857.. These results establish that execution-based validation provides a more reliable foundation for tool selection than semantic similarity alone, enabling more robust agent performance in real-world applications.
Authors: Sang Hun Kim, Jongmin Lee, Dongkyu Park, So Young Lee, Yosep Chong
Abstract: We propose a multi-agent framework for modeling artificial consciousness in large language models (LLMs), grounded in psychoanalytic theory. Our \textbf{Psychodynamic Model} simulates self-awareness, preconsciousness, and unconsciousness through agent interaction, guided by a Personalization Module combining fixed traits and dynamic needs. Using parameter-efficient fine-tuning on emotionally rich dialogues, the system was evaluated across eight personalized conditions. An LLM as a judge approach showed a 71.2\% preference for the fine-tuned model, with improved emotional depth and reduced output variance, demonstrating its potential for adaptive, personalized cognition.
Authors: Jusheng Zhang, Kaitong Cai, Yijia Fan, Ningyuan Liu, Keze Wang
Abstract: Multi-label image classification demands adaptive training strategies to navigate complex, evolving visual-semantic landscapes, yet conventional methods rely on static configurations that falter in dynamic settings. We propose MAT-Agent, a novel multi-agent framework that reimagines training as a collaborative, real-time optimization process. By deploying autonomous agents to dynamically tune data augmentation, optimizers, learning rates, and loss functions, MAT-Agent leverages non-stationary multi-armed bandit algorithms to balance exploration and exploitation, guided by a composite reward harmonizing accuracy, rare-class performance, and training stability. Enhanced with dual-rate exponential moving average smoothing and mixed-precision training, it ensures robustness and efficiency. Extensive experiments across Pascal VOC, COCO, and VG-256 demonstrate MAT-Agent's superiority: it achieves an mAP of 97.4 (vs. 96.2 for PAT-T), OF1 of 92.3, and CF1 of 91.4 on Pascal VOC; an mAP of 92.8 (vs. 92.0 for HSQ-CvN), OF1 of 88.2, and CF1 of 87.1 on COCO; and an mAP of 60.9, OF1 of 70.8, and CF1 of 61.1 on VG-256. With accelerated convergence and robust cross-domain generalization, MAT-Agent offers a scalable, intelligent solution for optimizing complex visual models, paving the way for adaptive deep learning advancements.
Authors: Waleed Razzaq, Yun-Bo Zhao
Abstract: Prognostic Health Management (PHM) systems monitor and predict equipment health. A key task is Remaining Useful Life (RUL) estimation, which predicts how long a component, such as a rolling element bearing, will operate before failure. Many RUL methods exist but often lack generalizability and robustness under changing operating conditions. This paper introduces CARLE, a hybrid AI framework that combines deep and shallow learning to address these challenges. CARLE uses Res-CNN and Res-LSTM blocks with multi-head attention and residual connections to capture spatial and temporal degradation patterns, and a Random Forest Regressor (RFR) for stable, accurate RUL prediction. A compact preprocessing pipeline applies Gaussian filtering for noise reduction and Continuous Wavelet Transform (CWT) for time-frequency feature extraction. We evaluate CARLE on the XJTU-SY and PRONOSTIA bearing datasets. Ablation studies measure each component's contribution, while noise and cross-domain experiments test robustness and generalization. Comparative results show CARLE outperforms several state-of-the-art methods, especially under dynamic conditions. Finally, we analyze model interpretability with LIME and SHAP to assess transparency and trustworthiness.
Authors: Alexandre G. Leclercq, S\'ebastien Bougleux, No\'emie N. Moreau, Alexis Desmonts, Romain H\'erault, Aur\'elien Corroyer-Dulmont
Abstract: Glioblastoma (GBM) is an aggressive primary brain tumor with a median survival of approximately 15 months. In clinical practice, the Stupp protocol serves as the standard first-line treatment. However, patients exhibit highly heterogeneous therapeutic responses which required at least two months before first visual impact can be observed, typically with MRI. Early prediction treatment response is crucial for advancing personalized medicine. Disease Progression Modeling (DPM) aims to capture the trajectory of disease evolution, while Treatment Response Prediction (TRP) focuses on assessing the impact of therapeutic interventions. Whereas most TRP approaches primarly rely on timeseries data, we consider the problem of early visual TRP as a slice-to-slice translation model generating post-treatment MRI from a pre-treatment MRI, thus reflecting the tumor evolution. To address this problem we propose a Latent Diffusion Model with a concatenation-based conditioning from the pre-treatment MRI and the tumor localization, and a classifier-free guidance to enhance generation quality using survival information, in particular post-treatment tumor evolution. Our model were trained and tested on a local dataset consisting of 140 GBM patients collected at Centre Fran\c{c}ois Baclesse. For each patient we collected pre and post T1-Gd MRI, tumor localization manually delineated in the pre-treatment MRI by medical experts, and survival information.
Authors: Yuze Sun, Wentao Luo, Yanfei Xiang, Jiancheng Pan, Jiahao Li, Quan Zhang, Xiaomeng Huang
Abstract: With the growing role of artificial intelligence in climate and weather research, efficient model training and inference are in high demand. Current models like FourCastNet and AI-GOMS depend heavily on GPUs, limiting hardware independence, especially for Chinese domestic hardware and frameworks. To address this issue, we present a framework for migrating large-scale atmospheric and oceanic models from PyTorch to MindSpore and optimizing for Chinese chips, and evaluating their performance against GPUs. The framework focuses on software-hardware adaptation, memory optimization, and parallelism. Furthermore, the model's performance is evaluated across multiple metrics, including training speed, inference speed, model accuracy, and energy efficiency, with comparisons against GPU-based implementations. Experimental results demonstrate that the migration and optimization process preserves the models' original accuracy while significantly reducing system dependencies and improving operational efficiency by leveraging Chinese chips as a viable alternative for scientific computing. This work provides valuable insights and practical guidance for leveraging Chinese domestic chips and frameworks in atmospheric and oceanic AI model development, offering a pathway toward greater technological independence.
Authors: Sungmin Cho, Sungbum Park, Insoo Oh
Abstract: In this work, we introduce MUSE (Model-based Uncertainty-aware Similarity Estimation), a training-free framework designed for model-based zero-shot 2D object detection and segmentation. MUSE leverages 2D multi-view templates rendered from 3D unseen objects and 2D object proposals extracted from input query images. In the embedding stage, it integrates class and patch embeddings, where the patch embeddings are normalized using generalized mean pooling (GeM) to capture both global and local representations efficiently. During the matching stage, MUSE employs a joint similarity metric that combines absolute and relative similarity scores, enhancing the robustness of matching under challenging scenarios. Finally, the similarity score is refined through an uncertainty-aware object prior that adjusts for proposal reliability. Without any additional training or fine-tuning, MUSE achieves state-of-the-art performance on the BOP Challenge 2025, ranking first across the Classic Core, H3, and Industrial tracks. These results demonstrate that MUSE offers a powerful and generalizable framework for zero-shot 2D object detection and segmentation.
Authors: Jian-wei Liu, Bing-rong Xu, Zhi-yan Song
Abstract: In this paper, the branches of recursive and recurrent neural networks are classified in detail according to the network structure, training objective function and learning algorithm implementation. They are roughly divided into three categories: The first category is General Recursive and Recurrent Neural Networks, including Basic Recursive and Recurrent Neural Networks, Long Short Term Memory Recursive and Recurrent Neural Networks, Convolutional Recursive and Recurrent Neural Networks, Differential Recursive and Recurrent Neural Networks, One-Layer Recursive and Recurrent Neural Networks, High-Order Recursive and Recurrent Neural Networks, Highway Networks, Multidimensional Recursive and Recurrent Neural Networks, Bidirectional Recursive and Recurrent Neural Networks; the second category is Structured Recursive and Recurrent Neural Networks, including Grid Recursive and Recurrent Neural Networks, Graph Recursive and Recurrent Neural Networks, Temporal Recursive and Recurrent Neural Networks, Lattice Recursive and Recurrent Neural Networks, Hierarchical Recursive and Recurrent Neural Networks, Tree Recursive and Recurrent Neural Networks; the third category is Other Recursive and Recurrent Neural Networks, including Array Long Short Term Memory, Nested and Stacked Recursive and Recurrent Neural Networks, Memory Recursive and Recurrent Neural Networks. Various networks cross each other and even rely on each other to form a complex network of relationships. In the context of the development and convergence of various networks, many complex sequence, speech and image problems are solved. After a detailed description of the principle and structure of the above model and model deformation, the research progress and application of each model are described, and finally the recursive and recurrent neural network models are prospected and summarized.
Authors: Tadesse K Bahiru, Natnael Tilahun Sinshaw, Teshager Hailemariam Moges, Dheeraj Kumar Singh
Abstract: Gender classification systems often inherit and amplify demographic imbalances in their training data. We first audit five widely used gender classification datasets, revealing that all suffer from significant intersectional underrepresentation. To measure the downstream impact of these flaws, we train identical MobileNetV2 classifiers on the two most balanced of these datasets, UTKFace and FairFace. Our fairness evaluation shows that even these models exhibit significant bias, misclassifying female faces at a higher rate than male faces and amplifying existing racial skew. To counter these data-induced biases, we construct BalancedFace, a new public dataset created by blending images from FairFace and UTKFace, supplemented with images from other collections to fill missing demographic gaps. It is engineered to equalize subgroup shares across 189 intersections of age, race, and gender using only real, unedited images. When a standard classifier is trained on BalancedFace, it reduces the maximum True Positive Rate gap across racial subgroups by over 50% and brings the average Disparate Impact score 63% closer to the ideal of 1.0 compared to the next-best dataset, all with a minimal loss of overall accuracy. These results underline the profound value of data-centric interventions and provide an openly available resource for fair gender classification research.
Authors: Jason Tsay, Zidane Wright, Gaodan Fang, Kiran Kate, Saurabh Jha, Yara Rizk
Abstract: Agentic systems interact with external systems by calling tools such as Python functions, REST API endpoints, or command line tools such as kubectl in Kubernetes. These tool calls often fail for various syntactic and semantic reasons. Some less obvious semantic errors can only be identified and resolved after analyzing the tool's response. To repair these errors, we develop a post-tool execution reflection component that combines large language model (LLM)-based reflection with domain-specific retrieval-augmented generation (RAG) using documents describing both the specific tool being called and troubleshooting documents related to the tool. For this paper, we focus on the use case of the kubectl command line tool to manage Kubernetes, a platform for orchestrating cluster applications. Through a larger empirical study and a smaller manual evaluation, we find that our RAG-based reflection will repair kubectl commands such that they are both more likely to successfully execute (pass rate) for 55% of our models evaluated and 36% more likely to correctly answer the user query on average. We find that troubleshooting documents improve pass rate compared to official documentation by an average of 10%.
Authors: Xiaoxu Xu, Xuexun Liu, Jinlong Li, Yitian Yuan, Qiudan Zhang, Lin Ma, Nicu Sebe, Xu Wang
Abstract: 3D weakly supervised semantic segmentation (3D WSSS) aims to achieve semantic segmentation by leveraging sparse or low-cost annotated data, significantly reducing reliance on dense point-wise annotations. Previous works mainly employ class activation maps or pre-trained vision-language models to address this challenge. However, the low quality of pseudo-labels and the insufficient exploitation of 3D geometric priors jointly create significant technical bottlenecks in developing high-performance 3D WSSS models. In this paper, we propose a simple yet effective 3D weakly supervised semantic segmentation method that integrates 3D geometric priors into a class-aware guidance mechanism to generate high-fidelity pseudo labels. Concretely, our designed methodology first employs Class-Aware Label Refinement module to generate more balanced and accurate pseudo labels for semantic categrories. This initial refinement stage focuses on enhancing label quality through category-specific optimization. Subsequently, the Geometry-Aware Label Refinement component is developed, which strategically integrates implicit 3D geometric constraints to effectively filter out low-confidence pseudo labels that fail to comply with geometric plausibility. Moreover, to address the challenge of extensive unlabeled regions, we propose a Label Update strategy that integrates Self-Training to propagate labels into these areas. This iterative process continuously enhances pseudo-label quality while expanding label coverage, ultimately fostering the development of high-performance 3D WSSS models. Comprehensive experimental validation reveals that our proposed methodology achieves state-of-the-art performance on both ScanNet and S3DIS benchmarks while demonstrating remarkable generalization capability in unsupervised settings, maintaining competitive accuracy through its robust design.
Authors: Yiheng Wang
Abstract: Intelligent reflecting surface (IRS) assisted unmanned aerial vehicle (UAV) systems provide a new paradigm for reconfigurable and flexible wireless communications. To enable more energy efficient and spectrum efficient IRS assisted UAV wireless communications, this paper introduces a novel IRS-assisted UAV enabled spectrum sharing system with orthogonal frequency division multiplexing (OFDM). The goal is to maximize the energy efficiency (EE) of the secondary network by jointly optimizing the beamforming, subcarrier allocation, IRS phase shifts, and the UAV trajectory subject to practical transmit power and passive reflection constraints as well as UAV physical limitations. A physically grounded propulsion-energy model is adopted, with its tight upper bound used to form a tractable EE lower bound for the spectrum sharing system. To handle highly non convex, time coupled optimization problems with a mixed continuous and discrete policy space, we develop a deep reinforcement learning (DRL) approach based on the actor critic framework. Extended experiments show the significant EE improvement of the proposed DRL-based approach compared to several benchmark schemes, thus demonstrating the effectiveness and robustness of the proposed approach with mobility.
Authors: Zheyuan Lin, Siqi Cai, Haizhou Li
Abstract: EEG-based person identification enables applications in security, personalized brain-computer interfaces (BCIs), and cognitive monitoring. However, existing techniques often rely on deep learning architectures at high computational cost, limiting their scope of applications. In this study, we propose a novel EEG person identification approach using spiking neural networks (SNNs) with a lightweight spiking transformer for efficiency and effectiveness. The proposed SNN model is capable of handling the temporal complexities inherent in EEG signals. On the EEG-Music Emotion Recognition Challenge dataset, the proposed model achieves 100% classification accuracy with less than 10% energy consumption of traditional deep neural networks. This study offers a promising direction for energy-efficient and high-performance BCIs. The source code is available at https://github.com/PatrickZLin/Decode-ListenerIdentity.
URLs: https://github.com/PatrickZLin/Decode-ListenerIdentity.
Authors: Hao Liu, Yiqing Dai, Haotian Tan, Yu Lei, Yujia Zhou, Zhen Wu
Abstract: Emotions guide human decisions, but whether large language models (LLMs) use emotion similarly remains unknown. We tested this using altruistic third-party punishment, where an observer incurs a personal cost to enforce fairness, a hallmark of human morality and often driven by negative emotion. In a large-scale comparison of 4,068 LLM agents with 1,159 adults across 796,100 decisions, LLMs used emotion to guide punishment, sometimes even more strongly than humans did: Unfairness elicited stronger negative emotion that led to more punishment; punishing unfairness produced more positive emotion than accepting; and critically, prompting self-reports of emotion causally increased punishment. However, mechanisms diverged: LLMs prioritized emotion over cost, enforcing norms in an almost all-or-none manner with reduced cost sensitivity, whereas humans balanced fairness and cost. Notably, reasoning models (o3-mini, DeepSeek-R1) were more cost-sensitive and closer to human behavior than foundation models (GPT-3.5, DeepSeek-V3), yet remained heavily emotion-driven. These findings provide the first causal evidence of emotion-guided moral decisions in LLMs and reveal deficits in cost calibration and nuanced fairness judgements, reminiscent of early-stage human responses. We propose that LLMs progress along a trajectory paralleling human development; future models should integrate emotion with context-sensitive reasoning to achieve human-like emotional intelligence.
Authors: Yizhuo Chen, Xin Liu, Ruijie Wang, Zheng Li, Pei Chen, Changlong Yu, Priyanka Nigam, Meng Jiang, Bing Yin
Abstract: Large language models (LLMs) achieve strong benchmark performance, yet user experiences remain inconsistent due to diverse preferences in style, tone, and reasoning mode. Nevertheless, existing alignment techniques such as reinforcement learning from human feedback (RLHF) or Direct Preference Optimization (DPO) largely optimize toward population-level averages and overlook individual variation. Naive personalization strategies like per-user fine-tuning are computationally prohibitive, and in-context approaches that prepend raw user signals often suffer from inefficiency and noise. To address these challenges, we propose POPI, a general framework that introduces a preference inference model to distill heterogeneous user signals into concise natural language summaries. These summaries act as transparent, compact, and transferable personalization representations that condition a shared generation model to produce personalized responses. POPI jointly optimizes both preference inference and personalized generation under a unified objective using reinforcement learning, ensuring summaries maximally encode useful preference information. Extensive experiments across four personalization benchmarks demonstrate that POPI consistently improves personalization accuracy while reducing context overhead by a large margin. Moreover, optimized summaries seamlessly transfer to frozen off-the-shelf LLMs, enabling plug-and-play personalization without weight updates.
Authors: Minfeng Qi, Zhongmin Cao, Qin Wang, Ningran Li, Tianqing Zhu
Abstract: Preprint repositories become central infrastructures for scholarly communication. Their expansion transforms how research is circulated and evaluated before journal publication. Generative large language models (LLMs) introduce a further potential disruption by altering how manuscripts are written. While speculation abounds, systematic evidence of whether and how LLMs reshape scientific publishing remains limited. This paper addresses the gap through a large-scale analysis of more than 2.1 million preprints spanning 2016--2025 (115 months) across four major repositories (i.e., arXiv, bioRxiv, medRxiv, SocArXiv). We introduce a multi-level analytical framework that integrates interrupted time-series models, collaboration and productivity metrics, linguistic profiling, and topic modeling to assess changes in volume, authorship, style, and disciplinary orientation. Our findings reveal that LLMs have accelerated submission and revision cycles, modestly increased linguistic complexity, and disproportionately expanded AI-related topics, while computationally intensive fields benefit more than others. These results show that LLMs act less as universal disruptors than as selective catalysts, amplifying existing strengths and widening disciplinary divides. By documenting these dynamics, the paper provides the first empirical foundation for evaluating the influence of generative AI on academic publishing and highlights the need for governance frameworks that preserve trust, fairness, and accountability in an AI-enabled research ecosystem.
Authors: Mohammad Abdul Rehman, Syed Imad Ali Shah, Abbas n=Anwar, Noor Islam
Abstract: Large Language Models (LLMs) can reason over natural-language inputs, but their role in intrusion detection without fine-tuning remains uncertain. This study evaluates a prompt-only approach on UNSW-NB15 by converting each network flow to a compact textual record and augmenting it with lightweight, domain-inspired boolean flags (asymmetry, burst rate, TTL irregularities, timer anomalies, rare service/state, short bursts). To reduce output drift and support measurement, the model is constrained to produce structured, grammar-valid responses, and a single decision threshold is calibrated on a small development split. We compare zero-shot, instruction-guided, and few-shot prompting to strong tabular and neural baselines under identical splits, reporting accuracy, precision, recall, F1, and macro scores. Empirically, unguided prompting is unreliable, while instructions plus flags substantially improve detection quality; adding calibrated scoring further stabilizes results. On a balanced subset of two hundred flows, a 7B instruction-tuned model with flags reaches macro-F1 near 0.78; a lighter 3B model with few-shot cues and calibration attains F1 near 0.68 on one thousand examples. As the evaluation set grows to two thousand flows, decision quality decreases, revealing sensitivity to coverage and prompting. Tabular baselines remain more stable and faster, yet the prompt-only pipeline requires no gradient training, produces readable artifacts, and adapts easily through instructions and flags. Contributions include a flow-to-text protocol with interpretable cues, a calibration method for thresholding, a systematic baseline comparison, and a reproducibility bundle with prompts, grammar, metrics, and figures.
Authors: Mohammad Abdul Rehman, Syed Imad Ali Shah, Abbas Anwar, Noor Islam
Abstract: The remarkable capabilities of Large Language Models (LLMs) in natural language understanding and generation have sparked interest in their potential for cybersecurity applications, including password guessing. In this study, we conduct an empirical investigation into the efficacy of pre-trained LLMs for password cracking using synthetic user profiles. Specifically, we evaluate the performance of state-of-the-art open-source LLMs such as TinyLLaMA, Falcon-RW-1B, and Flan-T5 by prompting them to generate plausible passwords based on structured user attributes (e.g., name, birthdate, hobbies). Our results, measured using Hit@1, Hit@5, and Hit@10 metrics under both plaintext and SHA-256 hash comparisons, reveal consistently poor performance, with all models achieving less than 1.5% accuracy at Hit@10. In contrast, traditional rule-based and combinator-based cracking methods demonstrate significantly higher success rates. Through detailed analysis and visualization, we identify key limitations in the generative reasoning of LLMs when applied to the domain-specific task of password guessing. Our findings suggest that, despite their linguistic prowess, current LLMs lack the domain adaptation and memorization capabilities required for effective password inference, especially in the absence of supervised fine-tuning on leaked password datasets. This study provides critical insights into the limitations of LLMs in adversarial contexts and lays the groundwork for future efforts in secure, privacy-preserving, and robust password modeling.
Authors: Hongyuan Liu, Xinyang Liu, Guosheng Hu
Abstract: The rapid advancement of Artificial Intelligence (AI) has created unprecedented demands for computational power, yet methods for evaluating the performance, efficiency, and environmental impact of deployed models remain fragmented. Current approaches often fail to provide a holistic view, making it difficult to compare and optimise systems across heterogeneous hardware, software stacks, and numeric precisions. To address this gap, we propose a unified and reproducible methodology for AI model inference that integrates computational and environmental metrics under realistic serving conditions. Our framework provides a pragmatic, carbon-aware evaluation by systematically measuring latency and throughput distributions, energy consumption, and location-adjusted carbon emissions, all while maintaining matched accuracy constraints for valid comparisons. We apply this methodology to multi-precision models across diverse hardware platforms, from data-centre accelerators like the GH200 to consumer-level GPUs such as the RTX 4090, running on mainstream software stacks including PyTorch, TensorRT, and ONNX Runtime. By systematically categorising these factors, our work establishes a rigorous benchmarking framework that produces decision-ready Pareto frontiers, clarifying the trade-offs between accuracy, latency, energy, and carbon. The accompanying open-source code enables independent verification and facilitates adoption, empowering researchers and practitioners to make evidence-based decisions for sustainable AI deployment.
Authors: Eilene Tomkins-Flanagan (Department of Cognitive Science, Carleton University), Mary A. Kelly (Department of Cognitive Science, Carleton University)
Abstract: Kanerva (2014) suggested that it would be possible to construct a complete Lisp out of a vector-symbolic architecture. We present the general form of a vector-symbolic representation of the five Lisp elementary functions, lambda expressions, and other auxiliary functions, found in the Lisp 1.5 specification McCarthy (1960), which is near minimal and sufficient for Turing-completeness. Our specific implementation uses holographic reduced representations Plate (1995), with a lookup table cleanup memory. Lisp, as all Turing-complete languages, is a Cartesian closed category, unusual in its proximity to the mathematical abstraction. We discuss the mathematics, the purpose, and the significance of demonstrating vector-symbolic architectures' Cartesian-closure, as well as the importance of explicitly including cleanup memories in the specification of the architecture.
Authors: Yunfei Liang
Abstract: Recent advances in deep learning have led to a surge of open-source models across diverse domains. While model merging offers a promising way to combine their strengths, existing approaches often suffer from parameter conflicts that degrade performance on domain-specific tasks. We propose MIN-Merging, a router-based framework that selectively merges the most important neurons to reduce such conflicts. Extensive experiments on Computer Vision(CV) and Natural Language Processing(NLP) benchmarks show that MIN-Merging achieves consistent gains on in-domain tasks while retaining the generalization ability of pretrained models on out-of-domain tasks. These results highlight its effectiveness as a practical solution to the parameter conflict problem in model merging.
Authors: Yisheng Zhong, Zhengbang Yang, Zhuangdi Zhu
Abstract: Large Language Models (LLMs) are increasingly integrated into real-world applications, raising concerns about privacy, security and the need to remove undesirable knowledge. Machine Unlearning has emerged as a promising solution, yet faces two key challenges: (1) practical unlearning needs are often continuous and heterogeneous, and (2) they involve decentralized, sensitive data with asymmetric access. These factors result in inter-domain and intra-domain interference, which further amplifies the dilemma of unbalanced forgetting and retaining performance. In response, we propose a federated unlearning approach for LLMs that is scalable and privacy preserving. Our method decouples unlearning and retention via task-specific adapter learning and employs a hierarchical merging strategy to mitigate conflicting objectives and enables robust, adaptable unlearning updates. Comprehensive experiments on benchmarks of WMDP, MUSE, and TOFU showed that our approach effectively handles heterogeneous unlearning requests while maintaining strong LLM utility compared with baseline methods.
Authors: Tao Bu, Qiangang Wang, Bowen Zeng, Hanwen Sun, Yunpeng Huang, Chun Cao, Jingwei Xu
Abstract: Transformer-based large language models (LLMs) have achieved remarkable success, yet their standard attention mechanism incurs quadratic computation and memory costs with respect to sequence length, posing a major bottleneck for long-context training. Prior work tackles this challenge along two directions: (1) kernel-level optimizations, which accelerate dense and sparse attention operators; and (2) module-level strategies, often referred to as distributed attention or context parallel training, which scale attention across multiple devices. However, systematic evaluation still remains limited: operator-level comparisons are often incomplete, while context parallel strategies are typically framework-specific, with unclear performance analysis across contexts. To address these gaps, we propose a unified benchmark that integrates representative attention kernels and context parallel mechanisms with a modular and extensible interface for evaluation. The benchmark evaluates methods along two critical dimensions: (1) attention mask patterns, which strongly affect efficiency, scalability, and usability, and (2) sequence length and distributed scale, which determine performance under extreme long-context training. Through comprehensive experiments on the cluster of up to 96 GPUs, our benchmark enables reproducible comparisons, highlights method-specific trade-offs, and provides practical guidance for designing and deploying attention mechanisms in long-context LLM training.
Authors: Shihao Ji, Zihui Song
Abstract: The Mixture of Experts (MoE) architecture enables the scaling of Large Language Models (LLMs) to trillions of parameters by activating a sparse subset of weights for each input, maintaining constant computational cost during inference. Concurrently, Low-Rank Adaptation (LoRA) has emerged as a dominant technique for parameter-efficiently fine-tuning LLMs on specialized tasks. In this work, we unify these two paradigms into a novel, end-to-end trainable framework named L-MoE: a Lightweight Mixture of LoRA Experts. L-MoE redefines MoE experts not as dense feed-forward networks, but as a collection of task-specialized, low-rank adapters. A lightweight gating network, trained jointly with the experts, learns to dynamically compose these LoRA adapters by computing a weighted average of their parameters for each input token. This composition is fully differentiable, allowing gradients from a standard auto-regressive language modeling objective to flow back through the entire architecture, simultaneously refining both the expert adapters and the routing strategy. This approach creates a highly parameter-efficient MoE model that is modular by design, allows for dynamic skill composition, and is trainable from end-to-end. We present the formal mathematical framework for L-MoE, detailing the differentiable routing mechanism and the joint optimization objective, thereby providing a new path toward building more efficient, scalable, and specialized language models.
Authors: Floris-Jan Willemsen, Niki van Stein, Ben van Werkhoven
Abstract: Automatic performance tuning (auto-tuning) is essential for optimizing high-performance applications, where vast and irregular parameter spaces make manual exploration infeasible. Traditionally, auto-tuning relies on well-established optimization algorithms such as evolutionary algorithms, annealing methods, or surrogate model-based optimizers to efficiently find near-optimal configurations. However, designing effective optimizers remains challenging, as no single method performs best across all tuning tasks. In this work, we explore a new paradigm: using large language models (LLMs) to automatically generate optimization algorithms tailored to auto-tuning problems. We introduce a framework that prompts LLMs with problem descriptions and search-space characteristics results to produce specialized optimization strategies, which are iteratively examined and improved. These generated algorithms are evaluated on four real-world auto-tuning applications across six hardware platforms and compared against the state-of-the-art in optimization algorithms of two contemporary auto-tuning frameworks. The evaluation demonstrates that providing additional application- and search space-specific information in the generation stage results in an average performance improvement of 30.7\% and 14.6\%, respectively. In addition, our results show that LLM-generated optimizers can rival, and in various cases outperform, existing human-designed algorithms, with our best-performing generated optimization algorithms achieving, on average, 72.4\% improvement over state-of-the-art optimizers for auto-tuning.
Authors: Kush Juvekar, Arghya Bhattacharya, Sai Khadloya, Utkarsh Saxena
Abstract: Large language models (LLMs) are entering legal workflows, yet we lack a jurisdiction-specific framework to assess their baseline competence therein. We use India's public legal examinations as a transparent proxy. Our multi-year benchmark assembles objective screens from top national and state exams and evaluates open and frontier LLMs under real-world exam conditions. To probe beyond multiple-choice questions, we also include a lawyer-graded, paired-blinded study of long-form answers from the Supreme Court's Advocate-on-Record exam. This is, to our knowledge, the first exam-grounded, India-specific yardstick for LLM court-readiness released with datasets and protocols. Our work shows that while frontier systems consistently clear historical cutoffs and often match or exceed recent top-scorer bands on objective exams, none surpasses the human topper on long-form reasoning. Grader notes converge on three reliability failure modes: procedural or format compliance, authority or citation discipline, and forum-appropriate voice and structure. These findings delineate where LLMs can assist (checks, cross-statute consistency, statute and precedent lookups) and where human leadership remains essential: forum-specific drafting and filing, procedural and relief strategy, reconciling authorities and exceptions, and ethical, accountable judgment.
Authors: Alex Acero, Daniel M. Jimenez-Gutierrez, Dario Pighin, Enrique Zuazua, Joaquin Del Rio, Xabi Uribe-Etxebarria
Abstract: Federated Learning (FL) enables collaborative decentralized training across multiple parties (nodes) while keeping raw data private. There are two main paradigms in FL: Horizontal FL (HFL), where all participant nodes share the same feature space but hold different samples, and Vertical FL (VFL), where participants hold complementary features for the same samples. While HFL is widely adopted, VFL is employed in domains where nodes hold complementary features about the same samples. Still, VFL presents a significant limitation: the vast number of communications required during training. This compromises privacy and security, and can lead to high energy consumption, and in some cases, make model training unfeasible due to the high number of communications. In this paper, we introduce Sherpa.ai Blind Vertical Federated Learning (SBVFL), a novel paradigm that leverages a distributed training mechanism enhanced for privacy and security. Decoupling the vast majority of node updates from the server dramatically reduces node-server communication. Experiments show that SBVFL reduces communication by ~99% compared to standard VFL while maintaining accuracy and robustness. Therefore, SBVFL enables practical, privacy-preserving VFL across sensitive domains, including healthcare, finance, manufacturing, aerospace, cybersecurity, and the defense industry.
Authors: Amirkia Rafiei Oskooei, Mehmet S. Aktas
Abstract: The proficiency of Large Language Models (LLMs) in processing structured data and adhering to syntactic rules is a capability that drives their widespread adoption but also makes them paradoxically vulnerable. In this paper, we investigate this vulnerability through BreakFun, a jailbreak methodology that weaponizes an LLM's adherence to structured schemas. BreakFun employs a three-part prompt that combines an innocent framing and a Chain-of-Thought distraction with a core "Trojan Schema"--a carefully crafted data structure that compels the model to generate harmful content, exploiting the LLM's strong tendency to follow structures and schemas. We demonstrate this vulnerability is highly transferable, achieving an average success rate of 89% across 13 foundational and proprietary models on JailbreakBench, and reaching a 100% Attack Success Rate (ASR) on several prominent models. A rigorous ablation study confirms this Trojan Schema is the attack's primary causal factor. To counter this, we introduce the Adversarial Prompt Deconstruction guardrail, a defense that utilizes a secondary LLM to perform a "Literal Transcription"--extracting all human-readable text to isolate and reveal the user's true harmful intent. Our proof-of-concept guardrail demonstrates high efficacy against the attack, validating that targeting the deceptive schema is a viable mitigation strategy. Our work provides a look into how an LLM's core strengths can be turned into critical weaknesses, offering a fresh perspective for building more robustly aligned models.
Authors: Sagnik Dakshit, Sushmita Sinha Roy
Abstract: Large Language Models (LLMs) are increasingly being used in education, yet their correctness alone does not capture the quality, reliability, or pedagogical validity of their problem-solving behavior, especially in mathematics, where multistep logic, symbolic reasoning, and conceptual clarity are critical. Conventional evaluation methods largely focus on final answer accuracy and overlook the reasoning process. To address this gap, we introduce a novel interpretability framework for analyzing LLM-generated solutions using undergraduate calculus problems as a representative domain. Our approach combines reasoning flow extraction and decomposing solutions into semantically labeled operations and concepts with prompt ablation analysis to assess input salience and output stability. Using structured metrics such as reasoning complexity, phrase sensitivity, and robustness, we evaluated the model behavior on real Calculus I to III university exams. Our findings revealed that LLMs often produce syntactically fluent yet conceptually flawed solutions, with reasoning patterns sensitive to prompt phrasing and input variation. This framework enables fine-grained diagnosis of reasoning failures, supports curriculum alignment, and informs the design of interpretable AI-assisted feedback tools. This is the first study to offer a structured, quantitative, and pedagogically grounded framework for interpreting LLM reasoning in mathematics education, laying the foundation for the transparent and responsible deployment of AI in STEM learning environments.
Authors: Monika Zamojska, Jaros{\l}aw A. Chudziak
Abstract: Simulating nuanced human social dynamics with Large Language Models (LLMs) remains a significant challenge, particularly in achieving psychological depth and consistent persona behavior crucial for high-fidelity training tools. This paper introduces TACLA (Transactional Analysis Contextual LLM-based Agents), a novel Multi-Agent architecture designed to overcome these limitations. TACLA integrates core principles of Transactional Analysis (TA) by modeling agents as an orchestrated system of distinct Parent, Adult, and Child ego states, each with its own pattern memory. An Orchestrator Agent prioritizes ego state activation based on contextual triggers and an agent's life script, ensuring psychologically authentic responses. Validated in an educational scenario, TACLA demonstrates realistic ego state shifts in Student Agents, effectively modeling conflict de-escalation and escalation based on different teacher intervention strategies. Evaluation shows high conversational credibility and confirms TACLA's capacity to create dynamic, psychologically-grounded social simulations, advancing the development of effective AI tools for education and beyond.
Authors: Rikard Vinge, Isabelle Wittmann, Jannik Schneider, Michael Marszalek, Luis Gilch, Thomas Brunschwiler, Conrad M Albrecht
Abstract: We introduce NeuCo-Bench, a novel benchmark framework for evaluating (lossy) neural compression and representation learning in the context of Earth Observation (EO). Our approach builds on fixed-size embeddings that act as compact, task-agnostic representations applicable to a broad range of downstream tasks. NeuCo-Bench comprises three core components: (i) an evaluation pipeline built around reusable embeddings, (ii) a new challenge mode with a hidden-task leaderboard designed to mitigate pretraining bias, and (iii) a scoring system that balances accuracy and stability. To support reproducibility, we release SSL4EO-S12-downstream, a curated multispectral, multitemporal EO dataset. We present initial results from a public challenge at the 2025 CVPR EARTHVISION workshop and conduct ablations with state-of-the-art foundation models. NeuCo-Bench provides a first step towards community-driven, standardized evaluation of neural embeddings for EO and beyond.
Authors: Hassan Gharoun, Mohammad Sadegh Khorshidi, Kasra Ranjbarigderi, Fang Chen, Amir H. Gandomi
Abstract: Despite extensive research on neural network calibration, existing methods typically apply global transformations that treat all predictions uniformly, overlooking the heterogeneous reliability of individual predictions. Furthermore, the relationship between improved calibration and effective uncertainty-aware decision-making remains largely unexplored. This paper presents a post-hoc calibration framework that leverages prediction reliability assessment to jointly enhance calibration quality and uncertainty-aware decision-making. The framework employs proximity-based conformal prediction to stratify calibration samples into putatively correct and putatively incorrect groups based on semantic similarity in feature space. A dual calibration strategy is then applied: standard isotonic regression calibrated confidence in putatively correct predictions, while underconfidence-regularized isotonic regression reduces confidence toward uniform distributions for putatively incorrect predictions, facilitating their identification for further investigations. A comprehensive evaluation is conducted using calibration metrics, uncertainty-aware performance measures, and empirical conformal coverage. Experiments on CIFAR-10 and CIFAR-100 with BiT and CoAtNet backbones show that the proposed method achieves lower confidently incorrect predictions, and competitive Expected Calibration Error compared with isotonic and focal-loss baselines. This work bridges calibration and uncertainty quantification through instance-level adaptivity, offering a practical post-hoc solution that requires no model retraining while improving both probability alignment and uncertainty-aware decision-making.
Authors: Michael James McCulloch
Abstract: The Free Energy Principle (FEP) states that self-organizing systems must minimize variational free energy to persist, but the path from principle to implementable algorithm has remained unclear. We present a constructive proof that the FEP can be realized through exact local credit assignment. The system decomposes gradient computation hierarchically: spatial credit via feedback alignment, temporal credit via eligibility traces, and structural credit via a Trophic Field Map (TFM) that estimates expected gradient magnitude for each connection block. We prove these mechanisms are exact at their respective levels and validate the central claim empirically: the TFM achieves 0.9693 Pearson correlation with oracle gradients. This exactness produces emergent capabilities including 98.6% retention after task interference, autonomous recovery from 75% structural damage, self-organized criticality (spectral radius p ~= 1.0$), and sample-efficient reinforcement learning on continuous control tasks without replay buffers. The architecture unifies Prigogine's dissipative structures, Friston's free energy minimization, and Hopfield's attractor dynamics, demonstrating that exact hierarchical inference over network topology can be implemented with local, biologically plausible rules.
Authors: Jinseong Park, Mijung Park
Abstract: Data unlearning aims to remove the influence of specific training samples from a trained model without requiring full retraining. Unlike concept unlearning, data unlearning in diffusion models remains underexplored and often suffers from quality degradation or incomplete forgetting. To address this, we first observe that most existing methods attempt to unlearn the samples at all diffusion time steps equally, leading to poor-quality generation. We argue that forgetting occurs disproportionately across time and frequency, depending on the model and scenarios. By selectively focusing on specific time-frequency ranges during training, we achieve samples with higher aesthetic quality and lower noise. We validate this improvement by applying our time-frequency selective approach to diverse settings, including gradient-based and preference optimization objectives, as well as both image-level and text-to-image tasks. Finally, to evaluate both deletion and quality of unlearned data samples, we propose a simple normalized version of SSCD. Together, our analysis and methods establish a clearer understanding of the unique challenges in data unlearning for diffusion models, providing practical strategies to improve both evaluation and unlearning performance.
Authors: Junlan Feng, Fanyu Meng, Chong Long, Pengyu Cong, Duqing Wang, Yan Zheng, Yuyao Zhang, Xuanchang Gao, Ye Yuan, Yunfei Ma, Zhijie Ren, Fan Yang, Na Wu, Di Jin, Chao Deng
Abstract: The hallucination and credibility concerns of large language models (LLMs) are global challenges that the industry is collectively addressing. Recently, a significant amount of advances have been made on post-training and inference techniques to mitigate these challenges. However, it is widely agreed that unsafe and hallucinations of LLMs intrinsically originate from pre-training, involving pre-training data and the next-token prediction learning mechanism. In this paper, we focus on enhancing pre-training data to improve the trustworthiness and safety of LLMs. Since the data is vast, it's almost impossible to entirely purge the data of factual errors, logical inconsistencies, or distributional biases. Moreover, the pre-training data lack grounding in real-world knowledge. Each piece of data is treated as a sequence of tokens rather than as a representation of a part of the world. To overcome these issues, we propose approaches to enhancing our pre-training data with its context in the world and increasing a substantial amount of data reflecting industrial scenarios. We argue that most source data are created by the authors for specific purposes in a certain spatial-temporal context. They have played a role in the real world. By incorporating related world context information, we aim to better anchor pre-training data within real-world scenarios, thereby reducing uncertainty in model training and enhancing the model's safety and trustworthiness. We refer to our Data with World Context as DWC. We continue pre-training an earlier checkpoint of JT-35B-Base with 1.5 trillion of DWC tokens. We introduce our post-training procedures to activate the potentials of DWC. Compared with the Qwen model of a similar scale, JT-Safe-35B achieves an average performance improvement of 1.79% on the Safety and Trustworthy evaluation benchmarks, while being pretrained with only 6.2 trillion tokens.
Authors: Tenghui Huang, Jinbo Wen, Jiawen Kang, Siyong Chen, Zhengtao Li, Tao Zhang, Dongning Liu, Jiacheng Wang, Chengjun Cai, Yinqiu Liu, Dusit Niyato
Abstract: Smart contracts play a significant role in automating blockchain services. Nevertheless, vulnerabilities in smart contracts pose serious threats to blockchain security. Currently, traditional detection methods primarily rely on static analysis and formal verification, which can result in high false-positive rates and poor scalability. Large Language Models (LLMs) have recently made significant progress in smart contract vulnerability detection. However, they still face challenges such as high inference costs and substantial computational overhead. In this paper, we propose ParaVul, a parallel LLM and retrieval-augmented framework to improve the reliability and accuracy of smart contract vulnerability detection. Specifically, we first develop Sparse Low-Rank Adaptation (SLoRA) for LLM fine-tuning. SLoRA introduces sparsification by incorporating a sparse matrix into quantized LoRA-based LLMs, thereby reducing computational overhead and resource requirements while enhancing their ability to understand vulnerability-related issues. We then construct a vulnerability contract dataset and develop a hybrid Retrieval-Augmented Generation (RAG) system that integrates dense retrieval with Best Matching 25 (BM25), assisting in verifying the results generated by the LLM. Furthermore, we propose a meta-learning model to fuse the outputs of the RAG system and the LLM, thereby generating the final detection results. After completing vulnerability detection, we design chain-of-thought prompts to guide LLMs to generate comprehensive vulnerability detection reports. Simulation results demonstrate the superiority of ParaVul, especially in terms of F1 scores, achieving 0.9398 for single-label detection and 0.9330 for multi-label detection.
Authors: Bishal Chhetri, B. V. Rathish Kumar
Abstract: The dynamics of tumor-immune interactions within a complex tumor microenvironment are typically modeled using a system of ordinary differential equations or partial differential equations. These models introduce some unknown parameters that need to be estimated accurately and efficiently from the limited and noisy experimental data. Moreover, due to the intricate biological complexity and limitations in experimental measurements, tumor-immune dynamics are not fully understood, and therefore, only partial knowledge of the underlying physics may be available, resulting in unknown or missing terms within the system of equations. In this study, we develop a cancer biology-informed neural network model(CBINN) to infer the unknown parameters in the system of equations as well as to discover the missing physics from sparse and noisy measurements. We test the performance of the CBINN model on three distinct nonlinear compartmental tumor-immune models and evaluate its robustness across multiple synthetic noise levels. By harnessing these highly nonlinear dynamics, our CBINN framework effectively estimates the unknown model parameters and uncovers the underlying physical laws or mathematical structures that govern these biological systems, even from scattered and noisy measurements. The models chosen here represent the dynamic patterns commonly observed in compartmental models of tumor-immune interactions, thereby validating the generalizability and efficacy of our methodology.
Authors: Keuntae Kim, Eunhye Jeong, Sehyeon Lee, Seohee Yoon, Yong Suk Choi
Abstract: Recent advances in enhancing the reasoning ability of large language models (LLMs) have been remarkably successful. LLMs trained with reinforcement learning (RL) for reasoning demonstrate strong performance in challenging tasks such as mathematics and coding, even with relatively small model sizes. However, despite these improvements in task accuracy, the assessment of creativity in LLM generations has been largely overlooked in reasoning tasks, in contrast to writing tasks. The lack of research on creativity assessment in reasoning primarily stems from two challenges: (1) the difficulty of defining the range of creativity, and (2) the necessity of human evaluation in the assessment process. To address these challenges, we propose CLAWS, a method that defines and classifies mathematical solutions into typical, creative, and hallucinated categories without human evaluation, by leveraging attention weights across prompt sections and output. CLAWS outperforms five existing white-box detection methods (Perplexity, Logit Entropy, Window Entropy, Hidden Score, and Attention Score) on five 7-8B math RL models (DeepSeek, Qwen, Mathstral, OpenMath2, and Oreal). We validate CLAWS on 4545 math problems collected from 181 math contests (AJHSME, AMC, AIME).
Authors: Shuodi Liu, Yingzhuo Liu, Zi Wang, Yusheng Wang, Huijia Wu, Liuyu Xiang, Zhaofeng He
Abstract: Large language models (LLMs) have demonstrated remarkable reasoning and planning capabilities, driving extensive research into task decomposition. Existing task decomposition methods focus primarily on memory, tool usage, and feedback mechanisms, achieving notable success in specific domains, but they often overlook the trade-off between performance and cost. In this study, we first conduct a comprehensive investigation on task decomposition, identifying six categorization schemes. Then, we perform an empirical analysis of three factors that influence the performance and cost of task decomposition: categories of approaches, characteristics of tasks, and configuration of decomposition and execution models, uncovering three critical insights and summarizing a set of practical principles. Building on this analysis, we propose the Select-Then-Decompose strategy, which establishes a closed-loop problem-solving process composed of three stages: selection, execution, and verification. This strategy dynamically selects the most suitable decomposition approach based on task characteristics and enhances the reliability of the results through a verification module. Comprehensive evaluations across multiple benchmarks show that the Select-Then-Decompose consistently lies on the Pareto frontier, demonstrating an optimal balance between performance and cost. Our code is publicly available at https://github.com/summervvind/Select-Then-Decompose.
Authors: Chenwei Tang, Jingyu Xing, Xinyu Liu, Wei Ju, Jiancheng Lv, Deng Xiong, Ziyue Qiao
Abstract: Reinforcement Learning (RL) has emerged as a powerful paradigm for advancing Large Language Models (LLMs), achieving remarkable performance in complex reasoning domains such as mathematics and code generation. However, current RL methods face a fundamental scalability bottleneck due to their heavy reliance on human-curated preference data or labeled datasets for reward modeling. To overcome this limitation, we explore RL on unlabeled data where models learn autonomously from continuous experience streams. The core challenge in this setting lies in reliable reward estimation without ground-truth supervision. Existing approaches like Test-Time RL address this through self-consistent consensus, but risk reinforcing incorrect pseudo-labels derived from majority voting. We introduce COMPASS (Composite Path and Answer Self-Scoring), a novel test-time reward mechanism that operates without external supervision. COMPASS integrates two complementary components: the Dual-Calibration Answer Reward (DCAR), which stabilizes training by establishing trustworthy pseudo-labels through confidence and credibility calibration, and the Decisive Path Reward (DPR), which directly optimizes the reasoning process quality beyond mere outcome supervision. By jointly reinforcing trustworthy consensus answers and highly decisive reasoning chains, the COMPASS systematically enhances the model's analytical capabilities. Extensive experiments show that COMPASS achieves significant and consistent performance gains across diverse reasoning tasks and model architectures, advancing a more scalable direction for LLMs to learn from continuous experience.
Authors: Yehor Tereshchenko, Mika H\"am\"al\"ainen
Abstract: This paper presents a comprehensive comparative analysis of Natural Language Processing (NLP) methods for automated toxicity detection in online gaming chats. Traditional machine learning models with embeddings, large language models (LLMs) with zero-shot and few-shot prompting, fine-tuned transformer models, and retrieval-augmented generation (RAG) approaches are evaluated. The evaluation framework assesses three critical dimensions: classification accuracy, processing speed, and computational costs. A hybrid moderation system architecture is proposed that optimizes human moderator workload through automated detection and incorporates continuous learning mechanisms. The experimental results demonstrate significant performance variations across methods, with fine-tuned DistilBERT achieving optimal accuracy-cost trade-offs. The findings provide empirical evidence for deploying cost-effective, efficient content moderation systems in dynamic online gaming environments.
Authors: George Ma, Anurag Koul, Qi Chen, Yawen Wu, Sachit Kuhar, Yu Yu, Aritra Sengupta, Varun Kumar, Murali Krishna Ramanathan
Abstract: Large Language Models (LLMs) excel at code-related tasks but often struggle in realistic software repositories, where project-specific APIs and cross-file dependencies are crucial. Retrieval-augmented methods mitigate this by injecting repository context at inference time. The low inference-time latency budget affects either retrieval quality or the added latency adversely impacts user experience. We address this limitation with SpecAgent, an agent that improves both latency and code-generation quality by proactively exploring repository files during indexing and constructing speculative context that anticipates future edits in each file. This indexing-time asynchrony allows thorough context computation, masking latency, and the speculative nature of the context improves code-generation quality. Additionally, we identify the problem of future context leakage in existing benchmarks, which can inflate reported performance. To address this, we construct a synthetic, leakage-free benchmark that enables a more realistic evaluation of our agent against baselines. Experiments show that SpecAgent consistently achieves absolute gains of 9-11% (48-58% relative) compared to the best-performing baselines, while significantly reducing inference latency.
Authors: He Du, Bowen Li, Aijun Yang, Siyang He, Qipeng Guo, Dacheng Tao
Abstract: Reliable verifiable data has become a key driver of capability gains in modern language models, enabling stable reinforcement learning with verifiable rewards and effective distillation that transfers competence across math, coding, and agentic tasks. Yet constructing generalizable synthetic verifiable data remains difficult due to hallucination-prone generation, and weak or trivial verification artifacts that fail to separate strong from weak solutions. Existing approaches often rely on task-specific heuristics or post-hoc filters that do not transfer across domains and lack a principled, universal evaluator of verifiability. In this work, we introduce an evolutionary, task-agnostic, strategy-guided, executably-checkable data synthesis framework that, from minimal seed supervision, jointly synthesizes problems, diverse candidate solutions, and verification artifacts, and iteratively discovers strategies via a consistency-based evaluator that enforces agreement between human-annotated and strategy-induced checks. This pipeline upgrades filtering into principled synthesis: it reliably assembles coherent, verifiable training instances and generalizes without domain-specific rules. Our experiments demonstrate the effectiveness of the proposed approach under both RLVR and model distillation training paradigms. The results show that training with our synthesized data yields significant improvements on both the LiveCodeBench and AgentBench-OS tasks, highlighting the robust generalization of our framework.
Authors: Xirui Zhang (papernest), Philippe de La Chevasnerie (papernest), Benoit Fabre (papernest)
Abstract: Extending Named Entity Recognition (NER) models to new PII entities in noisy spoken-language data is a common need. We find that jointly fine-tuning a BERT model on standard semantic entities (PER, LOC, ORG) and new pattern-based PII (EMAIL, PHONE) results in minimal degradation for original classes. We investigate this "peaceful coexistence," hypothesizing that the model uses independent semantic vs. morphological feature mechanisms. Using an incremental learning setup as a diagnostic tool, we measure semantic drift and find two key insights. First, the LOC (location) entity is uniquely vulnerable due to a representation overlap with new PII, as it shares pattern-like features (e.g., postal codes). Second, we identify a "reverse O-tag representation drift." The model, initially trained to map PII patterns to 'O', blocks new learning. This is resolved only by unfreezing the 'O' tag's classifier, allowing the background class to adapt and "release" these patterns. This work provides a mechanistic diagnosis of NER model adaptation, highlighting feature independence, representation overlap, and 'O' tag plasticity.
Authors: Andrew Bowne
Abstract: Unlike other military technologies driven by national security needs and developed with federal funding, AI is predominantly funded and advanced by commercial industry for civilian applications. However, there is a lack of understanding of the reasons commercial AI firms decide to work with the DoD or choose to abstain from the defence market. This thesis argues that the contract law and procurement framework are among the most significant obstacles. This research indicates that the commercial AI industry actually views the DoD as an attractive customer. However, this attraction is despite the obstacles presented by traditional contract law and procurement practices used to solicit and award contracts. Drawing on social exchange theory, this thesis introduces a theoretical framework, optimal buyer theory, to understand the factors that influence a commercial decision to engage with the DoD. Interviews from a sample of the participants explain why the AI industry holds such perceptions, opinions, and preferences about contracts generally and the DoD, specifically, in its role as a customer. This thesis concludes that commercial AI firms are attracted to contracts that are consistent with their business and technology considerations. Additionally, it develops best practices for leveraging existing contract law, primarily other transaction authority, to align contracting practices with commercial preferences and the machine learning development and deployment lifecycle.
Authors: Jiahao Tang, Henry Hengyuan Zhao, Lijian Wu, Yifei Tao, Dongxing Mao, Yang Wan, Jingru Tan, Min Zeng, Min Li, Alex Jinpeng Wang
Abstract: We introduce Chart2Code, a new benchmark for evaluating the chart understanding and code generation capabilities of large multimodal models (LMMs). Chart2Code is explicitly designed from a user-driven perspective, capturing diverse real-world scenarios and progressively increasing task difficulty. It consists of three levels: Level 1 (Chart Reproduction) reproduces charts from a reference figure and user query; Level 2 (Chart Editing) involves complex modifications such as changing chart types or adding elements; and Level 3 (Long-Table to Chart Generation) requires models to transform long, information-dense tables into faithful charts following user instructions. To our knowledge, this is the first hierarchical benchmark that reflects practical chart2code usage while systematically scaling task complexity. In total, Chart2Code contains 2,023 tasks across 22 chart types, paired with multi-level evaluation metrics that assess both code correctness and the visual fidelity of rendered charts. We benchmark 25 state-of-the-art (SoTA) LMMs, including both proprietary and the latest open-source models such as GPT-5, Qwen2.5-VL, InternVL3/3.5, MiMo-VL, and Seed-1.6-VL. Experimental results demonstrate that even the SoTA model GPT-5 averages only 0.57 on code-based evaluation and 0.22 on chart-quality assessment across the editing tasks, underscoring the difficulty of Chart2Code. We anticipate this benchmark will drive advances in multimodal reasoning and foster the development of more robust and general-purpose LMMs. Our code and data are available on Chart2Code.
Authors: Xiangbo Deng, Cheng Chen, Peng Yang
Abstract: Detecting regime shifts in chaotic time series is hard because observation-space signals are entangled with intrinsic variability. We propose Parameter--Space Changepoint Detection (Param--CPD), a two--stage framework that first amortizes Bayesian inference of governing parameters with a neural posterior estimator trained by simulation-based inference, and then applies a standard CPD algorithm to the resulting parameter trajectory. On Lorenz--63 with piecewise-constant parameters, Param--CPD improves F1, reduces localization error, and lowers false positives compared to observation--space baselines. We further verify identifiability and calibration of the inferred posteriors on stationary trajectories, explaining why parameter space offers a cleaner detection signal. Robustness analyses over tolerance, window length, and noise indicate consistent gains. Our results show that operating in a physically interpretable parameter space enables accurate and interpretable changepoint detection in nonlinear dynamical systems.
Authors: Haoyu Huang, Hong Ting Tsang, Jiaxin Bai, Xi Peng, Gong Zhang, Yangqiu Song
Abstract: Retrieval-augmented generation (RAG) has shown some success in augmenting large language models (LLMs) with external knowledge. However, as a non-parametric knowledge integration paradigm for LLMs, RAG methods heavily rely on external retrieval modules and the retrieved textual context prior. Especially for very large scale knowledge augmentation, they would introduce substantial inference latency due to expensive searches and much longer relevant context. In this paper, we propose a parametric knowledge integration method, called \textbf{AtlasKV}, a scalable, effective, and general way to augment LLMs with billion-scale knowledge graphs (KGs) (e.g. 1B triples) using very little GPU memory cost (e.g. less than 20GB VRAM). In AtlasKV, we introduce KG2KV and HiKVP to integrate KG triples into LLMs at scale with sub-linear time and memory complexity. It maintains strong knowledge grounding and generalization performance using the LLMs' inherent attention mechanism, and requires no external retrievers, long context priors, or retraining when adapting to new knowledge.
Authors: Jiale Zhao, Cong Liu, Yuxuan Zhang, Chengyue Gong, Zhenyi Zhang, Shifeng Jin, Zhenyu Liu
Abstract: Determining crystal structures from X-ray diffraction data is fundamental across diverse scientific fields, yet remains a significant challenge when data is limited to low resolution. While recent deep learning models have made breakthroughs in solving the crystallographic phase problem, the resulting low-resolution electron density maps are often ambiguous and difficult to interpret. To overcome this critical bottleneck, we introduce XDXD, to our knowledge, the first end-to-end deep learning framework to determine a complete atomic model directly from low-resolution single-crystal X-ray diffraction data. Our diffusion-based generative model bypasses the need for manual map interpretation, producing chemically plausible crystal structures conditioned on the diffraction pattern. We demonstrate that XDXD achieves a 70.4\% match rate for structures with data limited to 2.0~\AA{} resolution, with a root-mean-square error (RMSE) below 0.05. Evaluated on a benchmark of 24,000 experimental structures, our model proves to be robust and accurate. Furthermore, a case study on small peptides highlights the model's potential for extension to more complex systems, paving the way for automated structure solution in previously intractable cases.
Authors: Fu-Yun Wang, Han Zhang, Michael Gharbi, Hongsheng Li, Taesung Park
Abstract: We present UniRL-Zero, a unified reinforcement learning (RL) framework that boosts, multimodal language model understanding and reasoning, diffusion model multimedia generation, and their beneficial interaction capabilities within a unified model. Our work defines six scenarios for unified model reinforcement learning, providing systematic baselines for reinforcement learning of unified understanding and generation model. Our code is available at https://github.com/G-U-N/UniRL.
Authors: Ana En\'eriz Janeiro, Karina Pitombeira Pereira, Julio Mayol, Javier Crespo, Fernando Carballo, Juan B. Cabello, Manel Ramos-Casals, Bibiana P\'erez Corbacho, Juan Turnes
Abstract: AI is transforming medical practice and redefining the competencies that future healthcare professionals need to master. Despite international recommendations, the integration of AI into Medicine curricula in Spain had not been systematically evaluated until now. A cross-sectional study (July-September 2025) including Spanish universities offering the official degree in Medicine, according to the 'Register of Universities, Centers and Degrees (Registro de Universidades, Centros y T\'itulos RUCT)'. Curricula and publicly available institutional documentation were reviewed to identify courses and competencies related to AI in the 2025-2026 academic year. The analysis was performed using descriptive statistics. Of the 52 universities analyzed, ten (19.2%) offer specific AI courses, whereas 36 (69.2%) include no related content. Most of the identified courses are elective, with a credit load ranging from three to six ECTS, representing on average 1.17% of the total 360 credits of the degree. The University of Ja\'en is the only institution offering a compulsory course with AI content. The territorial analysis reveals marked disparities: Andalusia leads with 55.5% of its universities incorporating AI training, while several communities lack any initiative in this area. The integration of AI into the medical degree in Spain is incipient, fragmented, and uneven, with a low weight in ECTS. The limited training load and predominance of elective courses restrict the preparation of future physicians to practice in a healthcare environment increasingly mediated by AI. The findings support the establishment of minimum standards and national monitoring of indicators.
Authors: Stewart Slocum, Julian Minder, Cl\'ement Dumas, Henry Sleight, Ryan Greenblatt, Samuel Marks, Rowan Wang
Abstract: Knowledge editing techniques promise to implant new factual knowledge into large language models (LLMs). But do LLMs really believe these facts? We develop a framework to measure belief depth and use it to evaluate the success of knowledge editing techniques. We operationalize belief depth as the extent to which implanted knowledge 1) generalizes to related contexts (e.g. Fermi estimates several logical steps removed), 2) is robust to self-scrutiny and direct challenge, and 3) is represented similarly to genuine knowledge (as measured by linear probes). Our evaluations show that simple prompting and mechanistic editing techniques fail to implant knowledge deeply. In contrast, Synthetic Document Finetuning (SDF) - where models are trained on LLM-generated documents consistent with a fact - often succeeds at implanting beliefs that behave similarly to genuine knowledge. However, SDF's success is not universal, as implanted beliefs that contradict basic world knowledge are brittle and representationally distinct from genuine knowledge. Overall, our work introduces measurable criteria for belief depth and enables the rigorous evaluation necessary for deploying knowledge editing in real-world applications.
Authors: Grant McKenzie, Krzysztof Janowicz, Carsten Kessler
Abstract: Large-scale pre-trained machine learning models have reshaped our understanding of artificial intelligence across numerous domains, including our own field of geography. As with any new technology, trust has taken on an important role in this discussion. In this chapter, we examine the multifaceted concept of trust in foundation models, particularly within a geographic context. As reliance on these models increases and they become relied upon for critical decision-making, trust, while essential, has become a fractured concept. Here we categorize trust into three types: epistemic trust in the training data, operational trust in the model's functionality, and interpersonal trust in the model developers. Each type of trust brings with it unique implications for geographic applications. Topics such as cultural context, data heterogeneity, and spatial relationships are fundamental to the spatial sciences and play an important role in developing trust. The chapter continues with a discussion of the challenges posed by different forms of biases, the importance of transparency and explainability, and ethical responsibilities in model development. Finally, the novel perspective of geographic information scientists is emphasized with a call for further transparency, bias mitigation, and regionally-informed policies. Simply put, this chapter aims to provide a conceptual starting point for researchers, practitioners, and policy-makers to better understand trust in (generative) GeoAI.
Authors: Sridhar Mahadevan
Abstract: In this paper, we generalize Pearl's do-calculus to an Intuitionistic setting called $j$-stable causal inference inside a topos of sheaves. Our framework is an elaboration of the recently proposed framework of Topos Causal Models (TCMs), where causal interventions are defined as subobjects. We generalize the original setting of TCM using the Lawvere-Tierney topology on a topos, defined by a modal operator $j$ on the subobject classifier $\Omega$. We introduce $j$-do-calculus, where we replace global truth with local truth defined by Kripke-Joyal semantics, and formalize causal reasoning as structure-preserving morphisms that are stable along $j$-covers. $j$-do-calculus is a sound rule system whose premises and conclusions are formulas of the internal Intuitionistic logic of the causal topos. We define $j$-stability for conditional independences and interventional claims as local truth in the internal logic of the causal topos. We give three inference rules that mirror Pearl's insertion/deletion and action/observation exchange, and we prove soundness in the Kripke-Joyal semantics. A companion paper in preparation will describe how to estimate the required entities from data and instantiate $j$-do with standard discovery procedures (e.g., score-based and constraint-based methods), and will include experimental results on how to (i) form data-driven $j$-covers (via regime/section constructions), (ii) compute chartwise conditional independences after graph surgeries, and (iii) glue them to certify the premises of the $j$-do rules in practice
Authors: Neeladri Bhuiya, Madhav Aggarwal, Diptanshu Purwar
Abstract: Large Language Models (LLMs) are improving at an exceptional rate. With the advent of agentic workflows, multi-turn dialogue has become the de facto mode of interaction with LLMs for completing long and complex tasks. While LLM capabilities continue to improve, they remain increasingly susceptible to jailbreaking, especially in multi-turn scenarios where harmful intent can be subtly injected across the conversation to produce nefarious outcomes. While single-turn attacks have been extensively explored, adaptability, efficiency and effectiveness continue to remain key challenges for their multi-turn counterparts. To address these gaps, we present PLAGUE, a novel plug-and-play framework for designing multi-turn attacks inspired by lifelong-learning agents. PLAGUE dissects the lifetime of a multi-turn attack into three carefully designed phases (Primer, Planner and Finisher) that enable a systematic and information-rich exploration of the multi-turn attack family. Evaluations show that red-teaming agents designed using PLAGUE achieve state-of-the-art jailbreaking results, improving attack success rates (ASR) by more than 30% across leading models in a lesser or comparable query budget. Particularly, PLAGUE enables an ASR (based on StrongReject) of 81.4% on OpenAI's o3 and 67.3% on Claude's Opus 4.1, two models that are considered highly resistant to jailbreaks in safety literature. Our work offers tools and insights to understand the importance of plan initialization, context optimization and lifelong learning in crafting multi-turn attacks for a comprehensive model vulnerability evaluation.
Authors: Christopher A McClurg, Alan R Wagner
Abstract: We advance the understanding of robotic intervention in high-risk scenarios by examining their potential to distract and impede a school shooter. To evaluate this concept, we conducted a virtual reality study with 150 university participants role-playing as a school shooter. Within the simulation, an autonomous robot predicted the shooter's movements and positioned itself strategically to interfere and distract. The strategy the robot used to approach the shooter was manipulated -- either moving directly in front of the shooter (aggressive) or maintaining distance (passive) -- and the distraction method, ranging from no additional cues (low), to siren and lights (medium), to siren, lights, and smoke to impair visibility (high). An aggressive, high-distraction robot reduced the number of victims by 46.6% relative to a no-robot control. This outcome underscores both the potential of robotic intervention to enhance safety and the pressing ethical questions surrounding their use in school environments.
Authors: Jeff Shen, Francois Lanusse, Liam Holden Parker, Ollie Liu, Tom Hehir, Leopoldo Sarra, Lucas Meyer, Micah Bowles, Sebastian Wagner-Carena, Sebastian Wagner-Carena, Helen Qu, Siavash Golkar, Alberto Bietti, Hatim Bourfoune, Nathan Cassereau, Pierre Cornette, Keiya Hirashima, Geraud Krawezik, Ruben Ohana, Nicholas Lourie, Michael McCabe, Rudy Morel, Payel Mukhopadhyay, Mariel Pettee, Bruno R\'egaldo-Saint Blancard, Kyunghyun Cho, Miles Cranmer, Shirley Ho
Abstract: Sequential scientific data span many resolutions and domains, and unifying them into a common representation is a key step toward developing foundation models for the sciences. Astronomical spectra exemplify this challenge: massive surveys have collected millions of spectra across a wide range of wavelengths and resolutions, yet analyses remain fragmented across spectral domains (e.g., optical vs. infrared) and object types (e.g., stars vs. galaxies), limiting the ability to pool information across datasets. We present a deep learning model that jointly learns from heterogeneous spectra in a self-supervised manner. Our universal spectral tokenizer processes spectra from a variety of object types and resolutions directly on their native wavelength grids, producing intrinsically aligned, homogeneous, and physically meaningful representations that can be efficiently adapted to achieve competitive performance across a range of downstream tasks. For the first time, we demonstrate that a single model can unify spectral data across resolutions and domains, suggesting that our model can serve as a powerful building block for foundation models in astronomy -- and potentially extend to other scientific domains with heterogeneous sequential data, such as climate and healthcare.
Authors: Nishant Subramani, Alfredo Gomez, Mona Diab
Abstract: Modern language models are evaluated on large benchmarks, which are difficult to make sense of, especially for model selection. Looking at the raw evaluation numbers themselves using a model-centric lens, we propose SimBA, a three phase framework to Simplify Benchmark Analysis. The three phases of SimBA are: stalk, where we conduct dataset & model comparisons, prowl, where we discover a representative subset, and pounce, where we use the representative subset to predict performance on a held-out set of models. Applying SimBA to three popular LM benchmarks: HELM, MMLU, and BigBenchLite reveals that across all three benchmarks, datasets and models relate strongly to one another (stalk). We develop an representative set discovery algorithm which covers a benchmark using raw evaluation scores alone. Using our algorithm, we find that with 6.25% (1/16), 1.7% (1/58), and 28.4% (21/74) of the datasets for HELM, MMLU, and BigBenchLite respectively, we achieve coverage levels of at least 95% (prowl). Additionally, using just these representative subsets, we can both preserve model ranks and predict performance on a held-out set of models with near zero mean-squared error (pounce). Taken together, SimBA can help model developers improve efficiency during model training and dataset creators validate whether their newly created dataset differs from existing datasets in a benchmark. Our code is open source, available at https://github.com/nishantsubramani/simba.
Authors: Fengqing Jiang, Yichen Feng, Yuetai Li, Luyao Niu, Basel Alomair, Radha Poovendran
Abstract: The convergence of LLM-powered research assistants and AI-based peer review systems creates a critical vulnerability: fully automated publication loops where AI-generated research is evaluated by AI reviewers without human oversight. We investigate this through \textbf{BadScientist}, a framework that evaluates whether fabrication-oriented paper generation agents can deceive multi-model LLM review systems. Our generator employs presentation-manipulation strategies requiring no real experiments. We develop a rigorous evaluation framework with formal error guarantees (concentration bounds and calibration analysis), calibrated on real data. Our results reveal systematic vulnerabilities: fabricated papers achieve acceptance rates up to . Critically, we identify \textit{concern-acceptance conflict} -- reviewers frequently flag integrity issues yet assign acceptance-level scores. Our mitigation strategies show only marginal improvements, with detection accuracy barely exceeding random chance. Despite provably sound aggregation mathematics, integrity checking systematically fails, exposing fundamental limitations in current AI-driven review systems and underscoring the urgent need for defense-in-depth safeguards in scientific publishing.
Authors: Asim Mohamed, Martin Gubri
Abstract: Multilingual watermarking aims to make large language model (LLM) outputs traceable across languages, yet current methods still fall short. Despite claims of cross-lingual robustness, they are evaluated only on high-resource languages. We show that existing multilingual watermarking methods are not truly multilingual: they fail to remain robust under translation attacks in medium- and low-resource languages. We trace this failure to semantic clustering, which fails when the tokenizer vocabulary contains too few full-word tokens for a given language. To address this, we introduce STEAM, a back-translation-based detection method that restores watermark strength lost through translation. STEAM is compatible with any watermarking method, robust across different tokenizers and languages, non-invasive, and easily extendable to new languages. With average gains of +0.19 AUC and +40%p TPR@1% on 17 languages, STEAM provides a simple and robust path toward fairer watermarking across diverse languages.
Authors: Aymane Hassini
Abstract: The rise of Large Language Models (LLMs) has accelerated the long-standing goal of enabling natural language querying over complex, hybrid databases. Yet, this ambition exposes a dual challenge: reasoning jointly over structured, multi-relational schemas and the semantic content of linked unstructured assets. To overcome this, we present DynaQuery - a unified, self-adapting framework that serves as a practical blueprint for next-generation "Unbound Databases." At the heart of DynaQuery lies the Schema Introspection and Linking Engine (SILE), a novel systems primitive that elevates schema linking to a first-class query planning phase. We conduct a rigorous, multi-benchmark empirical evaluation of this structure-aware architecture against the prevalent unstructured Retrieval-Augmented Generation (RAG) paradigm. Our results demonstrate that the unstructured retrieval paradigm is architecturally susceptible to catastrophic contextual failures, such as SCHEMA_HALLUCINATION, leading to unreliable query generation. In contrast, our SILE-based design establishes a substantially more robust foundation, nearly eliminating this failure mode. Moreover, end-to-end validation on a complex, newly curated benchmark uncovers a key generalization principle: the transition from pure schema-awareness to holistic semantics-awareness. Taken together, our findings provide a validated architectural basis for developing natural language database interfaces that are robust, adaptable, and predictably consistent.
Authors: Ziyan Wang, Enmao Diao, Qi Le, Pu Wang, Minwoo Lee, Shu-ping Yeh, Evgeny Stupachenko, Hao Feng, Li Yang
Abstract: Structured pruning is a practical approach to deploying large language models (LLMs) efficiently, as it yields compact, hardware-friendly architectures. However, the dominant local paradigm is task-agnostic: by optimizing layer-wise reconstruction rather than task objectives, it tends to preserve perplexity or generic zero-shot behavior but fails to capitalize on modest task-specific calibration signals, often yielding limited downstream gains. We revisit global structured pruning and present GISP-Global Iterative Structured Pruning-a post-training method that removes attention heads and MLP channels using first-order, loss-based important weights aggregated at the structure level with block-wise normalization. An iterative schedule, rather than one-shot pruning, stabilizes accuracy at higher sparsity and mitigates perplexity collapse without requiring intermediate fine-tuning; the pruning trajectory also forms nested subnetworks that support a "prune-once, deploy-many" workflow. Furthermore, because importance is defined by a model-level loss, GISP naturally supports task-specific objectives; we instantiate perplexity for language modeling and a margin-based objective for decision-style tasks. Extensive experiments show that across Llama2-7B/13B, Llama3-8B, and Mistral-0.3-7B, GISP consistently lowers WikiText-2 perplexity and improves downstream accuracy, with especially strong gains at 40-50% sparsity; on DeepSeek-R1-Distill-Llama-3-8B with GSM8K, task-aligned calibration substantially boosts exact-match accuracy.
Authors: Roberto Brusnicki, David Pop, Yuan Gao, Mattia Piccinini, Johannes Betz
Abstract: Autonomous driving systems remain critically vulnerable to the long-tail of rare, out-of-distribution scenarios with semantic anomalies. While Vision Language Models (VLMs) offer promising reasoning capabilities, naive prompting approaches yield unreliable performance and depend on expensive proprietary models, limiting practical deployment. We introduce SAVANT (Semantic Analysis with Vision-Augmented Anomaly deTection), a structured reasoning framework that achieves high accuracy and recall in detecting anomalous driving scenarios from input images through layered scene analysis and a two-phase pipeline: structured scene description extraction followed by multi-modal evaluation. Our approach transforms VLM reasoning from ad-hoc prompting to systematic analysis across four semantic layers: Street, Infrastructure, Movable Objects, and Environment. SAVANT achieves 89.6% recall and 88.0% accuracy on real-world driving scenarios, significantly outperforming unstructured baselines. More importantly, we demonstrate that our structured framework enables a fine-tuned 7B parameter open-source model (Qwen2.5VL) to achieve 90.8% recall and 93.8% accuracy - surpassing all models evaluated while enabling local deployment at near-zero cost. By automatically labeling over 9,640 real-world images with high accuracy, SAVANT addresses the critical data scarcity problem in anomaly detection and provides a practical path toward reliable, accessible semantic monitoring for autonomous systems.
Authors: Harshini Suresha, Kavitha SH
Abstract: The red palm mite infestation has become a serious concern, particularly in regions with extensive palm cultivation, leading to reduced productivity and economic losses. Accurate and early identification of mite-infested plants is critical for effective management. The current study focuses on evaluating and comparing the ML model for classifying the affected plants and detecting the infestation. TriggerNet is a novel interpretable AI framework that integrates Grad-CAM, RISE, FullGrad, and TCAV to generate novel visual explanations for deep learning models in plant classification and disease detection. This study applies TriggerNet to address red palm mite (Raoiella indica) infestation, a major threat to palm cultivation and agricultural productivity. A diverse set of RGB images across 11 plant species, Arecanut, Date Palm, Bird of Paradise, Coconut Palm, Ginger, Citrus Tree, Palm Oil, Orchid, Banana Palm, Avocado Tree, and Cast Iron Plant was utilized for training and evaluation. Advanced deep learning models like CNN, EfficientNet, MobileNet, ViT, ResNet50, and InceptionV3, alongside machine learning classifiers such as Random Forest, SVM, and KNN, were employed for plant classification. For disease classification, all plants were categorized into four classes: Healthy, Yellow Spots, Reddish Bronzing, and Silk Webbing. Snorkel was used to efficiently label these disease classes by leveraging heuristic rules and patterns, reducing manual annotation time and improving dataset reliability.
Authors: Jay Phil Yoo, Kazuma Kobayashi, Souvik Chakraborty, Syed Bahauddin Alam
Abstract: Forecasting unobservable physical quantities from sparse, cross-domain sensor data is a central unsolved problem in scientific machine learning. Existing neural operators and large-scale forecasters rely on dense, co-located input-output fields and short temporal contexts, assumptions that fail in real-world systems where sensing and prediction occur on distinct physical manifolds and over long timescales. We introduce the Spatio-Temporal Operator Network (STONe), a non-autoregressive neural operator that learns a stable functional mapping between heterogeneous domains. By directly inferring high-altitude radiation dose fields from sparse ground-based neutron measurements, STONe demonstrates that operator learning can generalize beyond shared-domain settings. It defines a nonlinear operator between sensor and target manifolds that remains stable over long forecasting horizons without iterative recurrence. This challenges the conventional view that operator learning requires domain alignment or autoregressive propagation. Trained on 23 years of global neutron data, STONe achieves accurate 180-day forecasts with millisecond inference latency. The framework establishes a general principle for cross-domain operator inference, enabling real-time prediction of complex spatiotemporal fields in physics, climate, and energy systems.
Authors: Mahsa Valizadeh, Xiangjue Dong, Rui Tuo, James Caverlee
Abstract: Large Language Models (LLMs) excel at capturing latent semantics and contextual relationships across diverse modalities. However, in modeling user behavior from sequential interaction data, performance often suffers when such semantic context is limited or absent. We introduce LaMAR, a LLM-driven semantic enrichment framework designed to enrich such sequences automatically. LaMAR leverages LLMs in a few-shot setting to generate auxiliary contextual signals by inferring latent semantic aspects of a user's intent and item relationships from existing metadata. These generated signals, such as inferred usage scenarios, item intents, or thematic summaries, augment the original sequences with greater contextual depth. We demonstrate the utility of this generated resource by integrating it into benchmark sequential modeling tasks, where it consistently improves performance. Further analysis shows that LLM-generated signals exhibit high semantic novelty and diversity, enhancing the representational capacity of the downstream models. This work represents a new data-centric paradigm where LLMs serve as intelligent context generators, contributing a new method for the semi-automatic creation of training data and language resources.
Authors: Arman Behnam, Binghui Wang
Abstract: Causal representation learning in the anti-causal setting (labels cause features rather than the reverse) presents unique challenges requiring specialized approaches. We propose Anti-Causal Invariant Abstractions (ACIA), a novel measure-theoretic framework for anti-causal representation learning. ACIA employs a two-level design, low-level representations capture how labels generate observations, while high-level representations learn stable causal patterns across environment-specific variations. ACIA addresses key limitations of existing approaches by accommodating prefect and imperfect interventions through interventional kernels, eliminating dependency on explicit causal structures, handling high-dimensional data effectively, and providing theoretical guarantees for out-of-distribution generalization. Experiments on synthetic and real-world medical datasets demonstrate that ACIA consistently outperforms state-of-the-art methods in both accuracy and invariance metrics. Furthermore, our theoretical results establish tight bounds on performance gaps between training and unseen environments, confirming the efficacy of our approach for robust anti-causal learning.
Authors: Jiajun Fan, Tong Wei, Chaoran Cheng, Yuxin Chen, Ge Liu
Abstract: Balancing exploration and exploitation during reinforcement learning fine-tuning of generative models presents a critical challenge, as existing approaches rely on fixed divergence regularization that creates an inherent dilemma: strong regularization preserves model capabilities but limits reward optimization, while weak regularization enables greater alignment but risks instability or reward hacking. We introduce Adaptive Divergence Regularized Policy Optimization (ADRPO), which automatically adjusts regularization strength based on advantage estimates-reducing regularization for high-value samples while applying stronger regularization to poor samples, enabling policies to navigate between exploration and aggressive exploitation according to data quality. Our implementation with Wasserstein-2 regularization for flow matching generative models achieves remarkable results on text-to-image generation, achieving better semantic alignment and diversity than offline methods like DPO and online methods with fixed regularization like ORW-CFM-W2. ADRPO enables a 2B parameter SD3 model to surpass much larger models with 4.8B and 12B parameters in attribute binding, semantic consistency, artistic style transfer, and compositional control while maintaining generation diversity. ADRPO generalizes to KL-regularized fine-tuning of both text-only LLMs and multi-modal reasoning models, enhancing existing online RL methods like GRPO. In LLM fine-tuning, ADRPO demonstrates an emergent ability to escape local optima through active exploration, while in multi-modal audio reasoning, it outperforms GRPO through superior step-by-step reasoning, enabling a 7B model to outperform substantially larger commercial models including Gemini 2.5 Pro and GPT-4o Audio, offering an effective plug-and-play solution to the exploration-exploitation challenge across diverse generative architectures and modalities.
Authors: Wei-Jer Chang, Akshay Rangesh, Kevin Joseph, Matthew Strong, Masayoshi Tomizuka, Yihan Hu, Wei Zhan
Abstract: Developing autonomous vehicles (AVs) requires not only safety and efficiency, but also realistic, human-like behaviors that are socially aware and predictable. Achieving this requires sim agent policies that are human-like, fast, and scalable in multi-agent settings. Recent progress in imitation learning with large diffusion-based or tokenized models has shown that behaviors can be captured directly from human driving data, producing realistic policies. However, these models are computationally expensive, slow during inference, and struggle to adapt in reactive, closed-loop scenarios. In contrast, self-play reinforcement learning (RL) scales efficiently and naturally captures multi-agent interactions, but it often relies on heuristics and reward shaping, and the resulting policies can diverge from human norms. We propose SPACeR, a framework that leverages a pretrained tokenized autoregressive motion model as a centralized reference policy to guide decentralized self-play. The reference model provides likelihood rewards and KL divergence, anchoring policies to the human driving distribution while preserving RL scalability. Evaluated on the Waymo Sim Agents Challenge, our method achieves competitive performance with imitation-learned policies while being up to 10x faster at inference and 50x smaller in parameter size than large generative models. In addition, we demonstrate in closed-loop ego planning evaluation tasks that our sim agents can effectively measure planner quality with fast and scalable traffic simulation, establishing a new paradigm for testing autonomous driving policies.
Authors: Jiajun Fan, Chaoran Cheng, Shuaike Shen, Xiangxin Zhou, Ge Liu
Abstract: Flow-based generative models have shown remarkable success in text-to-image generation, yet fine-tuning them with intermediate feedback remains challenging, especially for continuous-time flow matching models. Most existing approaches solely learn from outcome rewards, struggling with the credit assignment problem. Alternative methods that attempt to learn a critic via direct regression on cumulative rewards often face training instabilities and model collapse in online settings. We present AC-Flow, a robust actor-critic framework that addresses these challenges through three key innovations: (1) reward shaping that provides well-normalized learning signals to enable stable intermediate value learning and gradient control, (2) a novel dual-stability mechanism that combines advantage clipping to prevent destructive policy updates with a warm-up phase that allows the critic to mature before influencing the actor, and (3) a scalable generalized critic weighting scheme that extends traditional reward-weighted methods while preserving model diversity through Wasserstein regularization. Through extensive experiments on Stable Diffusion 3, we demonstrate that AC-Flow achieves state-of-the-art performance in text-to-image alignment tasks and generalization to unseen human preference models. Our results demonstrate that even with a computationally efficient critic model, we can robustly finetune flow models without compromising generative quality, diversity, or stability.
Authors: Nadir Farhi
Abstract: In this work, we address the problem of determining reliable policies in reinforcement learning (RL), with a focus on optimization under uncertainty and the need for performance guarantees. While classical RL algorithms aim at maximizing the expected return, many real-world applications - such as routing, resource allocation, or sequential decision-making under risk - require strategies that ensure not only high average performance but also a guaranteed probability of success. To this end, we propose a novel formulation in which the objective is to maximize the probability that the cumulative return exceeds a prescribed threshold. We demonstrate that this reliable RL problem can be reformulated, via a state-augmented representation, into a standard RL problem, thereby allowing the use of existing RL and deep RL algorithms without the need for entirely new algorithmic frameworks. Theoretical results establish the equivalence of the two formulations and show that reliable strategies can be derived by appropriately adapting well-known methods such as Q-learning or Dueling Double DQN. To illustrate the practical relevance of the approach, we consider the problem of reliable routing, where the goal is not to minimize the expected travel time but rather to maximize the probability of reaching the destination within a given time budget. Numerical experiments confirm that the proposed formulation leads to policies that effectively balance efficiency and reliability, highlighting the potential of reliable RL for applications in stochastic and safety-critical environments.
Authors: Jiawei Zhang, Andrew Estornell, David D. Baek, Bo Li, Xiaojun Xu
Abstract: Large Language Models (LLMs) exhibit strong but shallow alignment: they directly refuse harmful queries when a refusal is expected at the very start of an assistant turn, yet this protection collapses once a harmful continuation is underway (either through the adversarial attacks or via harmful assistant-prefill attacks). This raises a fundamental question: Can the innate shallow alignment in LLMs be unlocked to ensure safety at arbitrary generation depths? To achieve this goal, we propose Any-Depth Alignment (ADA), an effective inference-time defense with negligible overhead. ADA is built based on our observation that alignment is concentrated in the assistant header tokens through repeated use in shallow-refusal training, and these tokens possess the model's strong alignment priors. By reintroducing these tokens mid-stream, ADA induces the model to reassess harmfulness and recover refusals at any point in generation. Across diverse open-source model families (Llama, Gemma, Mistral, Qwen, DeepSeek, and gpt-oss), ADA achieves robust safety performance without requiring any changes to the base model's parameters. It secures a near-100% refusal rate against challenging adversarial prefill attacks ranging from dozens to thousands of tokens. Furthermore, ADA reduces the average success rate of prominent adversarial prompt attacks (such as GCG, AutoDAN, PAIR, and TAP) to below 3%. This is all accomplished while preserving utility on benign tasks with minimal over-refusal. ADA maintains this resilience even after the base model undergoes subsequent instruction tuning (benign or adversarial).
Authors: Zaineh Abughazzah, Emna Baccour, Loay Ismail, Amr Mohamed, Mounir Hamdi
Abstract: The integration of Unmanned Aerial Vehicles (UAVs) into Open Radio Access Networks (O-RAN) enhances communication in disaster management and Search and Rescue (SAR) operations by ensuring connectivity when infrastructure fails. However, SAR scenarios demand stringent security and low-latency communication, as delays or breaches can compromise mission success. While UAVs serve as mobile relays, they introduce challenges in energy consumption and resource management, necessitating intelligent allocation strategies. Existing UAV-assisted O-RAN approaches often overlook the joint optimization of security, latency, and energy efficiency in dynamic environments. This paper proposes a novel Reinforcement Learning (RL)-based framework for dynamic resource allocation in UAV relays, explicitly addressing these trade-offs. Our approach formulates an optimization problem that integrates security-aware resource allocation, latency minimization, and energy efficiency, which is solved using RL. Unlike heuristic or static methods, our framework adapts in real-time to network dynamics, ensuring robust communication. Simulations demonstrate superior performance compared to heuristic baselines, achieving enhanced security and energy efficiency while maintaining ultra-low latency in SAR scenarios.
Authors: Connor Mattson, Varun Raveendra, Ellen Novoseller, Nicholas Waytowich, Vernon J. Lawhern, Daniel S. Brown
Abstract: Imitation Learning (IL) is a natural way for humans to teach robots, particularly when high-quality demonstrations are easy to obtain. While IL has been widely applied to single-robot settings, relatively few studies have addressed the extension of these methods to multi-agent systems, especially in settings where a single human must provide demonstrations to a team of collaborating robots. In this paper, we introduce and study Round-Robin Behavior Cloning (R2BC), a method that enables a single human operator to effectively train multi-robot systems through sequential, single-agent demonstrations. Our approach allows the human to teleoperate one agent at a time and incrementally teach multi-agent behavior to the entire system, without requiring demonstrations in the joint multi-agent action space. We show that R2BC methods match, and in some cases surpass, the performance of an oracle behavior cloning approach trained on privileged synchronized demonstrations across four multi-agent simulated tasks. Finally, we deploy R2BC on two physical robot tasks trained using real human demonstrations.
Authors: Rohan Choudhury, JungEun Kim, Jinhyung Park, Eunho Yang, L\'aszl\'o A. Jeni, Kris M. Kitani
Abstract: Vision Transformers (ViTs) partition input images into uniformly sized patches regardless of their content, resulting in long input sequence lengths for high-resolution images. We present Adaptive Patch Transformers (APT), which addresses this by using multiple different patch sizes within the same image. APT reduces the total number of input tokens by allocating larger patch sizes in more homogeneous areas and smaller patches in more complex ones. APT achieves a drastic speedup in ViT inference and training, increasing throughput by 40% on ViT-L and 50% on ViT-H while maintaining downstream performance, and can be applied to a previously fine-tuned ViT, converging in as little as 1 epoch. It also significantly reduces training and inference time without loss of performance in high-resolution dense visual tasks, achieving up to 30\% faster training and inference in visual QA, object detection, and semantic segmentation.
Authors: Nursultan Mamatov, Philipp Kellmeyer
Abstract: Accurate early prediction of in-hospital mortality in intensive care units (ICUs) is essential for timely clinical intervention and efficient resource allocation. This study develops and evaluates machine learning models that integrate both structured clinical data and unstructured textual information, specifically discharge summaries and radiology reports, from the MIMIC-IV database. We used LASSO and XGBoost for feature selection, followed by a multivariate logistic regression trained on the top features identified by both models. Incorporating textual features using TF-IDF and BERT embeddings significantly improved predictive performance. The final logistic regression model, which combined structured and textual input, achieved an AUC of 0.918, compared to 0.753 when using structured data alone, a relative improvement 22%. The analysis of the decision curve demonstrated a superior standardized net benefit in a wide range of threshold probabilities (0.2-0.8), confirming the clinical utility of the model. These results underscore the added prognostic value of unstructured clinical notes and support their integration into interpretable feature-driven risk prediction models for ICU patients.
Authors: Joeran Beel, Bela Gipp, Tobias Vente, Moritz Baumgart, Philipp Meister
Abstract: Recommender-systems research has accelerated model and evaluation advances, yet largely neglects automating the research process itself. We argue for a shift from narrow AutoRecSys tools -- focused on algorithm selection and hyper-parameter tuning -- to an Autonomous Recommender-Systems Research Lab (AutoRecLab) that integrates end-to-end automation: problem ideation, literature analysis, experimental design and execution, result interpretation, manuscript drafting, and provenance logging. Drawing on recent progress in automated science (e.g., multi-agent AI Scientist and AI Co-Scientist systems), we outline an agenda for the RecSys community: (1) build open AutoRecLab prototypes that combine LLM-driven ideation and reporting with automated experimentation; (2) establish benchmarks and competitions that evaluate agents on producing reproducible RecSys findings with minimal human input; (3) create review venues for transparently AI-generated submissions; (4) define standards for attribution and reproducibility via detailed research logs and metadata; and (5) foster interdisciplinary dialogue on ethics, governance, privacy, and fairness in autonomous research. Advancing this agenda can increase research throughput, surface non-obvious insights, and position RecSys to contribute to emerging Artificial Research Intelligence. We conclude with a call to organise a community retreat to coordinate next steps and co-author guidance for the responsible integration of automated research systems.
Authors: Dario Shariatian, Alain Durmus, Stefano Peluchetti
Abstract: We study discrete diffusion for language and other categorical data and focus on a common limitation of masked denoisers: reverse transitions typically factorize across positions, which can weaken joint structure and degrade quality in few-step generation. We propose \emph{Latent Discrete Diffusion Models} (LDDMs), which couple a masked discrete diffusion over tokens with a continuous diffusion over latent embeddings. The latent channel provides a softer signal and carries cross-token dependencies that help resolve ambiguities. We present two instantiations: (i) FUJI-LDDMs, which perform fully joint denoising of tokens and latents, and (ii) SEQ-LDDMs, which sequentially resolve the latent and then the discrete chain conditionally on it. For both variants we derive ELBO-style objectives and discuss design choices to learn informative latents yet amenable to diffusoin modeling. In experiments, LDDMs yield improvements on unconditional generation metrics as compared to state-of-the-art masked discrete diffusion baselines, and are effective at lower sampling budgets, where unmasking many tokens per step is desirable.
Authors: Xiangbo Gao, Tzu-Hsiang Lin, Ruojing Song, Yuheng Wu, Kuan-Ru Huang, Zicheng Jin, Fangzhou Lin, Shinan Liu, Zhengzhong Tu
Abstract: Collaborative driving systems leverage vehicle-to-everything (V2X) communication across multiple agents to enhance driving safety and efficiency. Traditional V2X systems take raw sensor data, neural features, or perception results as communication media, which face persistent challenges, including high bandwidth demands, semantic loss, and interoperability issues. Recent advances investigate natural language as a promising medium, which can provide semantic richness, decision-level reasoning, and human-machine interoperability at significantly lower bandwidth. Despite great promise, this paradigm shift also introduces new vulnerabilities within language communication, including message loss, hallucinations, semantic manipulation, and adversarial attacks. In this work, we present the first systematic study of full-stack safety and security issues in natural-language-based collaborative driving. Specifically, we develop a comprehensive taxonomy of attack strategies, including connection disruption, relay/replay interference, content spoofing, and multi-connection forgery. To mitigate these risks, we introduce an agentic defense pipeline, which we call SafeCoop, that integrates a semantic firewall, language-perception consistency checks, and multi-source consensus, enabled by an agentic transformation function for cross-frame spatial alignment. We systematically evaluate SafeCoop in closed-loop CARLA simulation across 32 critical scenarios, achieving 69.15% driving score improvement under malicious attacks and up to 67.32% F1 score for malicious detection. This study provides guidance for advancing research on safe, secure, and trustworthy language-driven collaboration in transportation systems. Our project page is https://xiangbogaobarry.github.io/SafeCoop.
Authors: Yohei Ikenoue, Hitomi Tashiro, Shigeru Kuroyanagi
Abstract: Prompt engineering is crucial for achieving reliable and effective outputs from large language models (LLMs), but its design requires specialized knowledge of prompting techniques and a deep understanding of target tasks. To address this challenge, we propose a novel method that adaptively selects task-appropriate prompting techniques based on users' abstract task descriptions and automatically generates high-quality prompts without relying on pre-existing templates or frameworks. The proposed method constructs a knowledge base that associates task clusters, characterized by semantic similarity across diverse tasks, with their corresponding prompting techniques. When users input task descriptions, the system assigns them to the most relevant task cluster and dynamically generates prompts by integrating techniques drawn from the knowledge base. An experimental evaluation of the proposed method on 23 tasks from BIG-Bench Extra Hard (BBEH) demonstrates superior performance compared with standard prompts and existing automatic prompt-generation tools, as measured by both arithmetic and harmonic mean scores. This research establishes a foundation for streamlining and standardizing prompt creation, enabling non-experts to effectively leverage LLMs.
Authors: Lukas Helff, Ruben H\"arle, Wolfgang Stammer, Felix Friedrich, Manuel Brack, Antonia W\"ust, Hikaru Shindo, Patrick Schramowski, Kristian Kersting
Abstract: Large language models (LLMs) excel at generating fluent text, but their internal reasoning remains opaque and difficult to control. Sparse autoencoders (SAEs) make hidden activations more interpretable by exposing latent features that often align with human concepts. Yet, these features are fragile and passive, offering no mechanism for systematic reasoning or model control. To address this, we introduce ActivationReasoning (AR), a framework that embeds explicit logical reasoning into the latent space of LLMs. It proceeds in three stages: (1) Finding latent representations, first latent concept representations are identified (e.g., via SAEs) and organized into a dictionary; (2) Activating propositions, at inference time AR detects activating concepts and maps them to logical propositions; and (3)Logical reasoning, applying logical rules over these propositions to infer higher-order structures, compose new concepts, and steer model behavior. We evaluate AR on multi-hop reasoning (PrOntoQA), abstraction and robustness to indirect concept cues (Rail2Country), reasoning over natural and diverse language (ProverQA), and context-sensitive safety (BeaverTails). Across all tasks, AR scales robustly with reasoning complexity, generalizes to abstract and context-sensitive tasks, and transfers across model backbones. These results demonstrate that grounding logical structure in latent activations not only improves transparency but also enables structured reasoning, reliable control, and alignment with desired behaviors, providing a path toward more reliable and auditable AI.
Authors: Fatima AlGhamdi, Omar Alharbi, Abdullah Aldwyish, Raied Aljadaany, Muhammad Kamran J Khan, Huda Alamri
Abstract: Detecting anomalies in crowded scenes is challenging due to severe inter-person occlusions and highly dynamic, context-dependent motion patterns. Existing approaches often struggle to adapt to varying crowd densities and lack interpretable anomaly indicators. To address these limitations, we introduce VelocityNet, a dual-pipeline framework that combines head detection and dense optical flow to extract person-specific velocities. Hierarchical clustering categorizes these velocities into semantic motion classes (halt, slow, normal, and fast), and a percentile-based anomaly scoring system measures deviations from learned normal patterns. Experiments demonstrate the effectiveness of our framework in real-time detection of diverse anomalous motion patterns within densely crowded environments.
Authors: Chengrun Li, Corentin Royer, Haozhe Luo, Bastian Wittmann, Xia Li, Ibrahim Hamamci, Sezgin Er, Anjany Sekuboyina, Bjoern Menze
Abstract: Most current medical vision language models struggle to jointly generate diagnostic text and pixel-level segmentation masks in response to complex visual questions. This represents a major limitation towards clinical application, as assistive systems that fail to provide both modalities simultaneously offer limited value to medical practitioners. To alleviate this limitation, we first introduce RadDiagSeg-D, a dataset combining abnormality detection, diagnosis, and multi-target segmentation into a unified and hierarchical task. RadDiagSeg-D covers multiple imaging modalities and is precisely designed to support the development of models that produce descriptive text and corresponding segmentation masks in tandem. Subsequently, we leverage the dataset to propose a novel vision-language model, RadDiagSeg-M, capable of joint abnormality detection, diagnosis, and flexible segmentation. RadDiagSeg-M provides highly informative and clinically useful outputs, effectively addressing the need to enrich contextual information for assistive diagnosis. Finally, we benchmark RadDiagSeg-M and showcase its strong performance across all components involved in the task of multi-target text-and-mask generation, establishing a robust and competitive baseline.
Authors: Yoshinari Fujinuma
Abstract: Large Language Models (LLMs) are commonly used as evaluators in various applications, but the reliability of the outcomes remains a challenge. One such challenge is using LLMs-as-judges for direct assessment, i.e., assigning scores from a specified range without any references. We first show that this challenge stems from LLM judge outputs being associated with score range bias, i.e., LLM judge outputs are highly sensitive to pre-defined score ranges, preventing the search for optimal score ranges. We also show that similar biases exist among models from the same family. We then mitigate this bias through contrastive decoding, achieving up to 11.3% relative improvement on average in Spearman correlation with human judgments across different score ranges.
Authors: Shruti Palaskar, Leon Gatys, Mona Abdelrahman, Mar Jacobo, Larry Lindsey, Rutika Moharir, Gunnar Lund, Yang Xu, Navid Shiee, Jeffrey Bigham, Charles Maalouf, Joseph Yitan Cheng
Abstract: Safety evaluation of multimodal foundation models often treats vision and language inputs separately, missing risks from joint interpretation where benign content becomes harmful in combination. Existing approaches also fail to distinguish clearly unsafe content from borderline cases, leading to problematic over-blocking or under-refusal of genuinely harmful content. We present Vision Language Safety Understanding (VLSU), a comprehensive framework to systematically evaluate multimodal safety through fine-grained severity classification and combinatorial analysis across 17 distinct safety patterns. Using a multi-stage pipeline with real-world images and human annotation, we construct a large-scale benchmark of 8,187 samples spanning 15 harm categories. Our evaluation of eleven state-of-the-art models reveals systematic joint understanding failures: while models achieve 90%-plus accuracy on clear unimodal safety signals, performance degrades substantially to 20-55% when joint image-text reasoning is required to determine the safety label. Most critically, 34% of errors in joint image-text safety classification occur despite correct classification of the individual modalities, further demonstrating absent compositional reasoning capabilities. Additionally, we find that models struggle to balance refusing unsafe content while still responding to borderline cases that deserve engagement. For example, we find that instruction framing can reduce the over-blocking rate on borderline content from 62.4% to 10.4% in Gemini-1.5, but only at the cost of under-refusing on unsafe content with refusal rate dropping from 90.8% to 53.9%. Overall, our framework exposes weaknesses in joint image-text understanding and alignment gaps in current models, and provides a critical test bed to enable the next milestones in research on robust vision-language safety.
Authors: Joseph Bejjani, Chase Van Amburg, Chengrui Wang, Chloe Huangyuan Su, Sarah M. Pratt, Yasin Mazloumi, Naeem Khoshnevis, Sham M. Kakade, Kiant\'e Brantley
Abstract: We explore how physical scale and population size shape the emergence of complex behaviors in open-ended ecological environments. In our setting, agents are unsupervised and have no explicit rewards or learning objectives but instead evolve over time according to reproduction, mutation, and natural selection. As they act, agents also shape their environment and the population around them in an ongoing dynamic ecology. Our goal is not to optimize a single high-performance policy, but instead to examine how behaviors emerge and evolve across large populations due to natural competition and environmental pressures. In an effort to discover how complex behaviors naturally emerge, we conduct experiments in large-scale worlds that reach populations of more than 60,000 individual agents, each with their own evolved neural network policy. We identify various emergent behaviors such as long-range resource extraction, vision-based foraging, and predation that arise under competitive and survival pressures. We examine how sensing modalities and environmental scale affect the emergence of these behaviors, finding that some appear only in sufficiently large environments and populations, with larger scales increasing behavioral stability and consistency. While there is a rich history of research in evolutionary settings, our scaling results provide promising new directions to explore ecology as an instrument of machine learning in an era of abundant computational resources. Experimental code is available at https://github.com/jbejjani2022/ecological-emergent-behavior.
URLs: https://github.com/jbejjani2022/ecological-emergent-behavior.
Authors: Jiangong Chen, Mingyu Zhu, Bin Li
Abstract: Mixed Reality (MR)-aided operation overlays digital objects on the physical world to provide a more immersive and intuitive operation process. A primary challenge is the precise and fast auto-verification of whether the user follows MR guidance by comparing frames before and after each operation. The pre-operation frame includes virtual guiding objects, while the post-operation frame contains physical counterparts. Existing approaches fall short of accounting for the discrepancies between physical and virtual objects due to imperfect 3D modeling or lighting estimation. In this paper, we propose EVER: an edge-assisted auto-verification system for mobile MR-aided operations. Unlike traditional frame-based similarity comparisons, EVER leverages the segmentation model and rendering pipeline adapted to the unique attributes of frames with physical pieces and those with their virtual counterparts; it adopts a threshold-based strategy using Intersection over Union (IoU) metrics for accurate auto-verification. To ensure fast auto-verification and low energy consumption, EVER offloads compute-intensive tasks to an edge server. Through comprehensive evaluations of public datasets and custom datasets with practical implementation, EVER achieves over 90% verification accuracy within 100 milliseconds (significantly faster than average human reaction time of approximately 273 milliseconds), while consuming only minimal additional computational resources and energy compared to a system without auto-verification.
Authors: Song Bian, Tao Yu, Shivaram Venkataraman, Youngsuk Park
Abstract: Scaling the number of parameters and the size of training data has proven to be an effective strategy for improving large language model (LLM) performance. Yet, as these models grow increasingly powerful and widely deployed, the cost of inference has become a pressing concern. Despite its importance, the trade-off between model accuracy and inference efficiency remains underexplored. In this work, we examine how key architectural factors, hidden size, the allocation of parameters between MLP and attention (mlp-to-attention ratio), and grouped-query attention (GQA), influence both inference cost and accuracy. We introduce a conditional scaling law that augments the Chinchilla framework with architectural information, along with a search framework for identifying architectures that are simultaneously inference-efficient and accurate. To validate our approach, we train more than 200 models spanning 80M to 3B parameters and 8B to 100B training tokens, and fit the proposed conditional scaling law. Our results show that the conditional scaling law reliably predicts optimal architectural choices and that the resulting models outperform existing open-source baselines. Under the same training budget, optimized architectures achieve up to 2.1% higher accuracy and 42% greater inference throughput compared to LLaMA-3.2.
Authors: Luis H. Chia
Abstract: Credit scoring models face a critical challenge: severe class imbalance, with default rates typically below 10%, which hampers model learning and predictive performance. While synthetic data augmentation techniques such as SMOTE and ADASYN have been proposed to address this issue, the optimal augmentation ratio remains unclear, with practitioners often defaulting to full balancing (1:1 ratio) without empirical justification. This study systematically evaluates 10 data augmentation scenarios using the Give Me Some Credit dataset (97,243 observations, 7% default rate), comparing SMOTE, BorderlineSMOTE, and ADASYN at different multiplication factors (1x, 2x, 3x). All models were trained using XGBoost and evaluated on a held-out test set of 29,173 real observations. Statistical significance was assessed using bootstrap testing with 1,000 iterations. Key findings reveal that ADASYN with 1x multiplication (doubling the minority class) achieved optimal performance with AUC of 0.6778 and Gini coefficient of 0.3557, representing statistically significant improvements of +0.77% and +3.00% respectively (p = 0.017, bootstrap test). Higher multiplication factors (2x and 3x) resulted in performance degradation, with 3x showing a -0.48% decrease in AUC, suggesting a "law of diminishing returns" for synthetic oversampling. The optimal class imbalance ratio was found to be 6.6:1 (majority:minority), contradicting the common practice of balancing to 1:1. This work provides the first empirical evidence of an optimal "sweet spot" for data augmentation in credit scoring, with practical guidelines for industry practitioners and researchers working with imbalanced datasets. While demonstrated on a single representative dataset, the methodology provides a reproducible framework for determining optimal augmentation ratios in other imbalanced domains.
Authors: Xiang Zhang, Suping Wu, Weibin Qiu, Zhaocheng Jin, Sheng Yang
Abstract: 3D human meshes show a natural hierarchical structure (like torso-limbs-fingers). But existing video-based 3D human mesh recovery methods usually learn mesh features in Euclidean space. It's hard to catch this hierarchical structure accurately. So wrong human meshes are reconstructed. To solve this problem, we propose a hyperbolic space learning method leveraging temporal motion prior for recovering 3D human meshes from videos. First, we design a temporal motion prior extraction module. This module extracts the temporal motion features from the input 3D pose sequences and image feature sequences respectively. Then it combines them into the temporal motion prior. In this way, it can strengthen the ability to express features in the temporal motion dimension. Since data representation in non-Euclidean space has been proved to effectively capture hierarchical relationships in real-world datasets (especially in hyperbolic space), we further design a hyperbolic space optimization learning strategy. This strategy uses the temporal motion prior information to assist learning, and uses 3D pose and pose motion information respectively in the hyperbolic space to optimize and learn the mesh features. Then, we combine the optimized results to get an accurate and smooth human mesh. Besides, to make the optimization learning process of human meshes in hyperbolic space stable and effective, we propose a hyperbolic mesh optimization loss. Extensive experimental results on large publicly available datasets indicate superiority in comparison with most state-of-the-art.
Authors: Tao Tao, Guanghui Zhu, Lang Guo, Hongyi Chen, Chunfeng Yuan, Yihua Huang
Abstract: Prompt Optimization has emerged as a crucial approach due to its capabilities in steering Large Language Models to solve various tasks. However, current works mainly rely on the random rewriting ability of LLMs, and the optimization process generally focus on specific influencing factors, which makes it easy to fall into local optimum. Besides, the performance of the optimized prompt is often unstable, which limits its transferability in different tasks. To address the above challenges, we propose $\textbf{DelvePO}$ ($\textbf{D}$irection-Guid$\textbf{e}$d Se$\textbf{l}$f-E$\textbf{v}$olving Framework for Fl$\textbf{e}$xible $\textbf{P}$rompt $\textbf{O}$ptimization), a task-agnostic framework to optimize prompts in self-evolve manner. In our framework, we decouple prompts into different components that can be used to explore the impact that different factors may have on various tasks. On this basis, we introduce working memory, through which LLMs can alleviate the deficiencies caused by their own uncertainties and further obtain key insights to guide the generation of new prompts. Extensive experiments conducted on different tasks covering various domains for both open- and closed-source LLMs, including DeepSeek-R1-Distill-Llama-8B, Qwen2.5-7B-Instruct and GPT-4o-mini. Experimental results show that DelvePO consistently outperforms previous SOTA methods under identical experimental settings, demonstrating its effectiveness and transferability across different tasks.
Authors: Xiaohan Qin, Xiaoxing Wang, Ning Liao, Junchi Yan
Abstract: Multi-Task Learning (MTL) enables a single model to learn multiple tasks simultaneously, leveraging knowledge transfer among tasks for enhanced generalization, and has been widely applied across various domains. However, task imbalance remains a major challenge in MTL. Although balancing the convergence speeds of different tasks is an effective approach to address this issue, it is highly challenging to accurately characterize the training dynamics and convergence speeds of multiple tasks within the complex MTL system. To this end, we attempt to analyze the training dynamics in MTL by leveraging Neural Tangent Kernel (NTK) theory and propose a new MTL method, NTKMTL. Specifically, we introduce an extended NTK matrix for MTL and adopt spectral analysis to balance the convergence speeds of multiple tasks, thereby mitigating task imbalance. Based on the approximation via shared representation, we further propose NTKMTL-SR, achieving training efficiency while maintaining competitive performance. Extensive experiments demonstrate that our methods achieve state-of-the-art performance across a wide range of benchmarks, including both multi-task supervised learning and multi-task reinforcement learning. Source code is available at https://github.com/jianke0604/NTKMTL.
Authors: Dechen Zhang, Junwei Su, Difan Zou
Abstract: The use of low-bit quantization has emerged as an indispensable technique for enabling the efficient training of large-scale models. Despite its widespread empirical success, a rigorous theoretical understanding of its impact on learning performance remains notably absent, even in the simplest linear regression setting. We present the first systematic theoretical study of this fundamental question, analyzing finite-step stochastic gradient descent (SGD) for high-dimensional linear regression under a comprehensive range of quantization targets: data, labels, parameters, activations, and gradients. Our novel analytical framework establishes precise algorithm-dependent and data-dependent excess risk bounds that characterize how different quantization affects learning: parameter, activation, and gradient quantization amplify noise during training; data quantization distorts the data spectrum; and data and label quantization introduce additional approximation and quantized error. Crucially, we prove that for multiplicative quantization (with input-dependent quantization step), this spectral distortion can be eliminated, and for additive quantization (with constant quantization step), a beneficial scaling effect with batch size emerges. Furthermore, for common polynomial-decay data spectra, we quantitatively compare the risks of multiplicative and additive quantization, drawing a parallel to the comparison between FP and integer quantization methods. Our theory provides a powerful lens to characterize how quantization shapes the learning dynamics of optimization algorithms, paving the way to further explore learning theory under practical hardware constraints.
Authors: Hua Su, Lei Zhang, Jin Zhao
Abstract: We introduce the Stable Physics-Informed Kernel Evolution (SPIKE) method for numerical computation of inviscid hyperbolic conservation laws. SPIKE resolves a fundamental paradox: how strong-form residual minimization can capture weak solutions containing discontinuities. SPIKE employs reproducing kernel representations with regularized parameter evolution, where Tikhonov regularization provides a smooth transition mechanism through shock formation, allowing the dynamics to traverse shock singularities. This approach automatically maintains conservation, tracks characteristics, and captures shocks satisfying Rankine-Hugoniot conditions within a unified framework requiring no explicit shock detection or artificial viscosity. Numerical validation across scalar and vector-valued conservation laws confirms the method's effectiveness.
Authors: Xiang Zhang, Suping Wu, Sheng Yang
Abstract: Existing 3D human mesh recovery methods often fail to fully exploit the latent information (e.g., human motion, shape alignment), leading to issues with limb misalignment and insufficient local details in the reconstructed human mesh (especially in complex scenes). Furthermore, the performance improvement gained by modelling mesh vertices and pose node interactions using attention mechanisms comes at a high computational cost. To address these issues, we propose a two-stage network for human mesh recovery based on latent information and low dimensional learning. Specifically, the first stage of the network fully excavates global (e.g., the overall shape alignment) and local (e.g., textures, detail) information from the low and high-frequency components of image features and aggregates this information into a hybrid latent frequency domain feature. This strategy effectively extracts latent information. Subsequently, utilizing extracted hybrid latent frequency domain features collaborates to enhance 2D poses to 3D learning. In the second stage, with the assistance of hybrid latent features, we model the interaction learning between the rough 3D human mesh template and the 3D pose, optimizing the pose and shape of the human mesh. Unlike existing mesh pose interaction methods, we design a low-dimensional mesh pose interaction method through dimensionality reduction and parallel optimization that significantly reduces computational costs without sacrificing reconstruction accuracy. Extensive experimental results on large publicly available datasets indicate superiority compared to the most state-of-the-art.
Authors: Xueyi Chen, Keda Tao, Kele Shao, Huan Wang
Abstract: Unlike offline processing, streaming video vision-language models face two fundamental constraints: causality and accumulation. Causality prevents access to future frames that offline methods exploit, while accumulation causes tokens to grow unbounded, creating efficiency bottlenecks. However, existing approaches only regulate post-LLM kv-cache, leaving costly pre-LLM prefill unchanged. We introduce StreamingTOM, a training-free, plug-and-play two-stage framework that addresses both pre-LLM and post-LLM bottlenecks with predictable latency. Causal Temporal Reduction imposes a fixed per-frame budget and selects tokens based on adjacent-frame changes and token saliency, drastically reducing per-frame prefill cost by processing only a compact subset of visual tokens per frame instead of all visual tokens. Online Quantized Memory stores tokens in 4-bit format, retrieves relevant groups on demand, and dequantizes them, keeping the active kv-cache bounded regardless of stream length. Experiments demonstrate our method achieves $15.7\times$ kv-cache compression, $1.2\times$ lower peak memory and $2\times$ faster TTFT compared to prior SOTA. StreamingTOM maintains state-of-the-art accuracy among training-free methods with an average of $63.8\%$ on offline benchmarks and $55.8\%/3.7$ on RVS. These results highlight the practical benefits of our two-stage approach for efficient streaming video understanding with bounded growth.
Authors: Yanhong Li, Zixuan Lan, Jiawei Zhou
Abstract: Large language models (LLMs) and their multimodal variants can now process visual inputs, including images of text. This raises an intriguing question: can we compress textual inputs by feeding them as images to reduce token usage while preserving performance? In this paper, we show that visual text representations are a practical and surprisingly effective form of input compression for decoder LLMs. We exploit the idea of rendering long text inputs as a single image and provide it directly to the model. This leads to dramatically reduced number of decoder tokens required, offering a new form of input compression. Through experiments on two distinct benchmarks RULER (long-context retrieval) and CNN/DailyMail (document summarization) we demonstrate that this text-as-image method yields substantial token savings (often nearly half) without degrading task performance.
Authors: Lei Li, Xiao Zhou, Yingying Zhang, Xian Wu
Abstract: Medical question answering (QA) requires extensive access to domain-specific knowledge. A promising direction is to enhance large language models (LLMs) with external knowledge retrieved from medical corpora or parametric knowledge stored in model parameters. Existing approaches typically fall into two categories: Retrieval-Augmented Generation (RAG), which grounds model reasoning on externally retrieved evidence, and Generation-Augmented Generation (GAG), which depends solely on the models internal knowledge to generate contextual documents. However, RAG often suffers from noisy or incomplete retrieval, while GAG is vulnerable to hallucinated or inaccurate information due to unconstrained generation. Both issues can mislead reasoning and undermine answer reliability. To address these challenges, we propose MedRGAG, a unified retrieval-generation augmented framework that seamlessly integrates external and parametric knowledge for medical QA. MedRGAG comprises two key modules: Knowledge-Guided Context Completion (KGCC), which directs the generator to produce background documents that complement the missing knowledge revealed by retrieval; and Knowledge-Aware Document Selection (KADS), which adaptively selects an optimal combination of retrieved and generated documents to form concise yet comprehensive evidence for answer generation. Extensive experiments on five medical QA benchmarks demonstrate that MedRGAG achieves a 12.5% improvement over MedRAG and a 4.5% gain over MedGENIE, highlighting the effectiveness of unifying retrieval and generation for knowledge-intensive reasoning. Our code and data are publicly available at https://anonymous.4open.science/r/MedRGAG
Authors: Brady Bhalla, Honglu Fan, Nancy Chen, Tony Yue YU
Abstract: We investigate how embedding dimension affects the emergence of an internal "world model" in a transformer trained with reinforcement learning to perform bubble-sort-style adjacent swaps. Models achieve high accuracy even with very small embedding dimensions, but larger dimensions yield more faithful, consistent, and robust internal representations. In particular, higher embedding dimensions strengthen the formation of structured internal representation and lead to better interpretability. After hundreds of experiments, we observe two consistent mechanisms: (1) the last row of the attention weight matrix monotonically encodes the global ordering of tokens; and (2) the selected transposition aligns with the largest adjacent difference of these encoded values. Our results provide quantitative evidence that transformers build structured internal world models and that model size improves representation quality in addition to end performance. We release our metrics and analyses, which can be used to probe similar algorithmic tasks.
Authors: Chengshu Li, Mengdi Xu, Arpit Bahety, Hang Yin, Yunfan Jiang, Huang Huang, Josiah Wong, Sujay Garlanka, Cem Gokmen, Ruohan Zhang, Weiyu Liu, Jiajun Wu, Roberto Mart\'in-Mart\'in, Li Fei-Fei
Abstract: Imitation learning from large-scale, diverse human demonstrations has proven effective for training robots, but collecting such data is costly and time-consuming. This challenge is amplified for multi-step bimanual mobile manipulation, where humans must teleoperate both a mobile base and two high-degree-of-freedom arms. Prior automated data generation frameworks have addressed static bimanual manipulation by augmenting a few human demonstrations in simulation, but they fall short for mobile settings due to two key challenges: (1) determining base placement to ensure reachability, and (2) positioning the camera to provide sufficient visibility for visuomotor policies. To address these issues, we introduce MoMaGen, which formulates data generation as a constrained optimization problem that enforces hard constraints (e.g., reachability) while balancing soft constraints (e.g., visibility during navigation). This formulation generalizes prior approaches and provides a principled foundation for future methods. We evaluate MoMaGen on four multi-step bimanual mobile manipulation tasks and show that it generates significantly more diverse datasets than existing methods. Leveraging this diversity, MoMaGen can train successful imitation learning policies from a single source demonstration, and these policies can be fine-tuned with as few as 40 real-world demonstrations to achieve deployment on physical robotic hardware. More details are available at our project page: momagen.github.io.
Authors: Zhong Li, Qi Huang, Yuxuan Zhu, Lincen Yang, Mohammad Mohammadi Amiri, Niki van Stein, Matthijs van Leeuwen
Abstract: We introduce Time-Conditioned Contraction Matching (TCCM), a novel method for semi-supervised anomaly detection in tabular data. TCCM is inspired by flow matching, a recent generative modeling framework that learns velocity fields between probability distributions and has shown strong performance compared to diffusion models and generative adversarial networks. Instead of directly applying flow matching as originally formulated, TCCM builds on its core idea -- learning velocity fields between distributions -- but simplifies the framework by predicting a time-conditioned contraction vector toward a fixed target (the origin) at each sampled time step. This design offers three key advantages: (1) a lightweight and scalable training objective that removes the need for solving ordinary differential equations during training and inference; (2) an efficient scoring strategy called one time-step deviation, which quantifies deviation from expected contraction behavior in a single forward pass, addressing the inference bottleneck of existing continuous-time models such as DTE (a diffusion-based model with leading anomaly detection accuracy but heavy inference cost); and (3) explainability and provable robustness, as the learned velocity field operates directly in input space, making the anomaly score inherently feature-wise attributable; moreover, the score function is Lipschitz-continuous with respect to the input, providing theoretical guarantees under small perturbations. Extensive experiments on the ADBench benchmark show that TCCM strikes a favorable balance between detection accuracy and inference cost, outperforming state-of-the-art methods -- especially on high-dimensional and large-scale datasets. The source code is available at our GitHub repository.
Authors: Alexandros Ntagkas, Chairi Kiourt, Konstantinos Chatzilygeroudis
Abstract: State-of-the-art perceptive Reinforcement Learning controllers for legged robots either (i) impose oscillator or IK-based gait priors that constrain the action space, add bias to the policy optimization and reduce adaptability across robot morphologies, or (ii) operate "blind", which struggle to anticipate hind-leg terrain, and are brittle to noise. In this paper, we propose Phase-Guided Terrain Traversal (PGTT), a perception-aware deep-RL approach that overcomes these limitations by enforcing gait structure purely through reward shaping, thereby reducing inductive bias in policy learning compared to oscillator/IK-conditioned action priors. PGTT encodes per-leg phase as a cubic Hermite spline that adapts swing height to local heightmap statistics and adds a swing- phase contact penalty, while the policy acts directly in joint space supporting morphology-agnostic deployment. Trained in MuJoCo (MJX) on procedurally generated stair-like terrains with curriculum and domain randomization, PGTT achieves the highest success under push disturbances (median +7.5% vs. the next best method) and on discrete obstacles (+9%), with comparable velocity tracking, and converging to an effective policy roughly 2x faster than strong end-to-end baselines. We validate PGTT on a Unitree Go2 using a real-time LiDAR elevation-to-heightmap pipeline, and we report preliminary results on ANYmal-C obtained with the same hyperparameters. These findings indicate that terrain-adaptive, phase-guided reward shaping is a simple and general mechanism for robust perceptive locomotion across platforms.
Authors: Giorgio Piras, Qi Zhao, Fabio Brau, Maura Pintor, Christian Wressnegger, Battista Biggio
Abstract: Adversarial pruning methods have emerged as a powerful tool for compressing neural networks while preserving robustness against adversarial attacks. These methods typically follow a three-step pipeline: (i) pretrain a robust model, (ii) select a binary mask for weight pruning, and (iii) finetune the pruned model. To select the binary mask, these methods minimize a robust loss by assigning an importance score to each weight, and then keep the weights with the highest scores. However, this score-space optimization can lead to sharp local minima in the robust loss landscape and, in turn, to an unstable mask selection, reducing the robustness of adversarial pruning methods. To overcome this issue, we propose a novel plug-in method for adversarial pruning, termed Score-space Sharpness-aware Adversarial Pruning (S2AP). Through our method, we introduce the concept of score-space sharpness minimization, which operates during the mask search by perturbing importance scores and minimizing the corresponding robust loss. Extensive experiments across various datasets, models, and sparsity levels demonstrate that S2AP effectively minimizes sharpness in score space, stabilizing the mask selection, and ultimately improving the robustness of adversarial pruning methods.
Authors: ChangSu Choi, Hoyun Song, Dongyeon Kim, WooHyeon Jung, Minkyung Cho, Sunjin Park, NohHyeob Bae, Seona Yu, KyungTae Lim
Abstract: Distilling the tool-using capabilities of large language models (LLMs) into smaller, more efficient small language models (SLMs) is a key challenge for their practical application. The predominant approach, supervised fine-tuning (SFT), suffers from poor generalization as it trains models to imitate a static set of teacher trajectories rather than learn a robust methodology. While reinforcement learning (RL) offers an alternative, the standard RL using sparse rewards fails to effectively guide SLMs, causing them to struggle with inefficient exploration and adopt suboptimal strategies. To address these distinct challenges, we propose MENTOR, a framework that synergistically combines RL with teacher-guided distillation. Instead of simple imitation, MENTOR employs an RL-based process to learn a more generalizable policy through exploration. In addition, to solve the problem of reward sparsity, it uses a teacher's reference trajectory to construct a dense, composite teacher-guided reward that provides fine-grained guidance. Extensive experiments demonstrate that MENTOR significantly improves the cross-domain generalization and strategic competence of SLMs compared to both SFT and standard sparse-reward RL baselines.
Authors: Mst Jannatun Ferdous, Masum Billah, Joy Karmoker, Mohd Ruhul Ameen, Akif Islam, Md. Omar Faruqe
Abstract: This paper presents an automated system for cricket video analysis that leverages deep learning techniques to extract wicket-taking deliveries, detect cricket balls, and model ball trajectories. The system employs the YOLOv8 architecture for pitch and ball detection, combined with optical character recognition (OCR) for scorecard extraction to identify wicket-taking moments. Through comprehensive image preprocessing, including grayscale transformation, power transformation, and morphological operations, the system achieves robust text extraction from video frames. The pitch detection model achieved 99.5% mean Average Precision at 50% IoU (mAP50) with a precision of 0.999, while the ball detection model using transfer learning attained 99.18% mAP50 with 0.968 precision and 0.978 recall. The system enables trajectory modeling on detected pitches, providing data-driven insights for identifying batting weaknesses. Experimental results on multiple cricket match videos demonstrate the effectiveness of this approach for automated cricket analytics, offering significant potential for coaching and strategic decision-making.
Authors: Miao Zhang, Junpeng Li, ChangChun HUa, Yana Yang
Abstract: Weakly supervised learning often operates with coarse aggregate signals rather than instance labels. We study a setting where each training example is an $n$-tuple containing exactly m positives, while only the count m per tuple is observed. This NTMP (N-tuple with M positives) supervision arises in, e.g., image classification with region proposals and multi-instance measurements. We show that tuple counts admit a trainable unbiased risk estimator (URE) by linking the tuple-generation process to latent instance marginals. Starting from fixed (n,m), we derive a closed-form URE and extend it to variable tuple sizes, variable counts, and their combination. Identification holds whenever the effective mixing rate is separated from the class prior. We establish generalization bounds via Rademacher complexity and prove statistical consistency with standard rates under mild regularity assumptions. To improve finite-sample stability, we introduce simple ReLU corrections to the URE that preserve asymptotic correctness. Across benchmarks converted to NTMP tasks, the approach consistently outperforms representative weak-supervision baselines and yields favorable precision-recall and F1 trade-offs. It remains robust under class-prior imbalance and across diverse tuple configurations, demonstrating that count-only supervision can be exploited effectively through a theoretically grounded and practically stable objective.
Authors: Rahul Soundrarajan, Claudio Fiandrino, Michele Polese, Salvatore D'Oro, Leonardo Bonati, Tommaso Melodia
Abstract: Open RAN introduces a flexible, cloud-based architecture for the Radio Access Network (RAN), enabling Artificial Intelligence (AI)/Machine Learning (ML)-driven automation across heterogeneous, multi-vendor deployments. While EXplainable Artificial Intelligence (XAI) helps mitigate the opacity of AI models, explainability alone does not guarantee reliable network operations. In this article, we propose a lightweight verification approach based on interpretable models to validate the behavior of Deep Reinforcement Learning (DRL) agents for RAN slicing and scheduling in Open RAN. Specifically, we use Decision Tree (DT)-based verifiers to perform near-real-time consistency checks at runtime, which would be otherwise unfeasible with computationally expensive state-of-the-art verifiers. We analyze the landscape of XAI and AI verification, propose a scalable architectural integration, and demonstrate feasibility with a DT-based slice-verifier. We also outline future challenges to ensure trustworthy AI adoption in Open RAN.
Authors: Alexander Bentkamp, Jasmin Blanchette, Matthias Hetzenberger, Uwe Waldmann
Abstract: The $\lambda$-superposition calculus is a successful approach to proving higher-order formulas. However, some parts of the calculus are extremely explosive, notably due to the higher-order unifier enumeration and the functional extensionality axiom. In the present work, we introduce an "optimistic" version of $\lambda$-superposition that addresses these two issues. Specifically, our new calculus delays explosive unification problems using constraints stored along with the clauses, and it applies functional extensionality in a more targeted way. The calculus is sound and refutationally complete with respect to a Henkin semantics. We have yet to implement it in a prover, but examples suggest that it will outperform, or at least usefully complement, the original $\lambda$-superposition calculus.
Authors: Zhiwei Hao, Jianyuan Guo, Li Shen, Kai Han, Yehui Tang, Han Hu, Yunhe Wang
Abstract: Recent advancements in vision transformers (ViTs) have demonstrated that larger models often achieve superior performance. However, training these models remains computationally intensive and costly. To address this challenge, we introduce ScaleNet, an efficient approach for scaling ViT models. Unlike conventional training from scratch, ScaleNet facilitates rapid model expansion with negligible increases in parameters, building on existing pretrained models. This offers a cost-effective solution for scaling up ViTs. Specifically, ScaleNet achieves model expansion by inserting additional layers into pretrained ViTs, utilizing layer-wise weight sharing to maintain parameters efficiency. Each added layer shares its parameter tensor with a corresponding layer from the pretrained model. To mitigate potential performance degradation due to shared weights, ScaleNet introduces a small set of adjustment parameters for each layer. These adjustment parameters are implemented through parallel adapter modules, ensuring that each instance of the shared parameter tensor remains distinct and optimized for its specific function. Experiments on the ImageNet-1K dataset demonstrate that ScaleNet enables efficient expansion of ViT models. With a 2$\times$ depth-scaled DeiT-Base model, ScaleNet achieves a 7.42% accuracy improvement over training from scratch while requiring only one-third of the training epochs, highlighting its efficiency in scaling ViTs. Beyond image classification, our method shows significant potential for application in downstream vision areas, as evidenced by the validation in object detection task.
Authors: Yuanhe Guo, Linxi Xie, Zhuoran Chen, Kangrui Yu, Ryan Po, Guandao Yang, Gordon Wetztein, Hongyi Wen
Abstract: We introduce ImageGem, a dataset for studying generative models that understand fine-grained individual preferences. We posit that a key challenge hindering the development of such a generative model is the lack of in-the-wild and fine-grained user preference annotations. Our dataset features real-world interaction data from 57K users, who collectively have built 242K customized LoRAs, written 3M text prompts, and created 5M generated images. With user preference annotations from our dataset, we were able to train better preference alignment models. In addition, leveraging individual user preference, we investigated the performance of retrieval models and a vision-language model on personalized image retrieval and generative model recommendation. Finally, we propose an end-to-end framework for editing customized diffusion models in a latent weight space to align with individual user preferences. Our results demonstrate that the ImageGem dataset enables, for the first time, a new paradigm for generative model personalization.
Authors: Tong Liu, Zhiwei Fan, Guanyan Peng, Haodan Zhang, Yucheng Zhang, Zhen Wang, Pengjin Xie, Liang Liu
Abstract: Short video streaming has become a dominant paradigm in digital media, characterized by rapid swiping interactions and diverse media content. A key technical challenge is designing an effective preloading strategy that dynamically selects and prioritizes download tasks from an evolving playlist, balancing Quality of Experience (QoE) and bandwidth efficiency under practical commercial constraints. However, real world analysis reveals critical limitations of existing approaches: (1) insufficient adaptation of download task sizes to dynamic conditions, and (2) watch time prediction models that are difficult to deploy reliably at scale. In this paper, we propose DeLoad, a novel preloading framework that addresses these issues by introducing dynamic task sizing and a practical, multi dimensional watch time estimation method. Additionally, a Deep Reinforcement Learning (DRL) enhanced agent is trained to optimize the download range decisions adaptively. Extensive evaluations conducted on an offline testing platform, leveraging massive real world network data, demonstrate that DeLoad achieves significant improvements in QoE metrics (34.4% to 87.4% gain). Furthermore, after deployment on a large scale commercial short video platform, DeLoad has increased overall user watch time by 0.09% while simultaneously reducing rebuffering events and 3.76% bandwidth consumption.
Authors: Yili Wang, Tairan Huang, Changlong He, Qiutong Li, Jianliang Gao
Abstract: Heterogeneous temporal graphs (HTGs) are ubiquitous data structures in the real world. Recently, to enhance representation learning on HTGs, numerous attention-based neural networks have been proposed. Despite these successes, existing methods rely on a decoupled temporal and spatial learning paradigm, which weakens interactions of spatio-temporal information and leads to a high model complexity. To bridge this gap, we propose a novel learning paradigm for HTGs called Simple and Efficient Heterogeneous Temporal Graph N}eural Network (SE-HTGNN). Specifically, we innovatively integrate temporal modeling into spatial learning via a novel dynamic attention mechanism, which retains attention information from historical graph snapshots to guide subsequent attention computation, thereby improving the overall discriminative representations learning of HTGs. Additionally, to comprehensively and adaptively understand HTGs, we leverage large language models to prompt SE-HTGNN, enabling the model to capture the implicit properties of node types as prior knowledge. Extensive experiments demonstrate that SE-HTGNN achieves up to 10x speed-up over the state-of-the-art and latest baseline while maintaining the best forecasting accuracy.
Authors: Xue Jiang, Yihong Dong, Mengyang Liu, Hongyi Deng, Tian Wang, Yongding Tao, Rongyu Cao, Binhua Li, Zhi Jin, Wenpin Jiao, Fei Huang, Yongbin Li, Ge Li
Abstract: While Large Language Models (LLMs) excel at code generation by learning from vast code corpora, a fundamental semantic gap remains between their training on textual patterns and the goal of functional correctness, which is governed by formal execution semantics. Reinforcement Learning with Verifiable Rewards (RLVR) approaches attempt to bridge this gap using outcome rewards from executing test cases. However, solely relying on binary pass/fail signals is inefficient for establishing a well-aligned connection between the textual representation of code and its execution semantics, especially for subtle logical errors within the code. In this paper, we propose CodeRL+, a novel approach that integrates execution semantics alignment into the RLVR training pipeline for code generation. CodeRL+ enables the model to infer variable-level execution trajectory, providing a direct learning signal of execution semantics. CodeRL+ can construct execution semantics alignment directly using existing on-policy rollouts and integrates seamlessly with various RL algorithms. Extensive experiments demonstrate that CodeRL+ outperforms post-training baselines (including RLVR and Distillation), achieving a 4.6% average relative improvement in pass@1. CodeRL+ generalizes effectively to other coding tasks, yielding 15.5% and 4.4% higher accuracy on code-reasoning and test-output-generation benchmarks, respectively. CodeRL+ shows strong applicability across diverse RL algorithms and LLMs. Furthermore, probe analyses provide compelling evidence that CodeRL+ strengthens the alignment between code's textual representations and its underlying execution semantics.
Authors: Yuya Sasaki
Abstract: Graph neural networks (GNNs) are powerful tools for learning from graph-structured data but often produce biased predictions with respect to sensitive attributes. Fairness-aware GNNs have been actively studied for mitigating biased predictions. However, no prior studies have evaluated fairness-aware GNNs on knowledge graphs, which are one of the most important graphs in many applications, such as recommender systems. Therefore, we introduce a benchmarking study on knowledge graphs. We generate new graphs from three knowledge graphs, YAGO, DBpedia, and Wikidata, that are significantly larger than the existing graph datasets used in fairness studies. We benchmark inprocessing and preprocessing methods in different GNN backbones and early stopping conditions. We find several key insights: (i) knowledge graphs show different trends from existing datasets; clearer trade-offs between prediction accuracy and fairness metrics than other graphs in fairness-aware GNNs, (ii) the performance is largely affected by not only fairness-aware GNN methods but also GNN backbones and early stopping conditions, and (iii) preprocessing methods often improve fairness metrics, while inprocessing methods improve prediction accuracy.
Authors: Kangzhong Wang, Zitong Shen, Youqian Zhang, Michael MK Cheung, Xiapu Luo, Grace Ngai, Eugene Yujun Fu
Abstract: Phone scams remain a pervasive threat to both personal safety and financial security worldwide. Recent advances in large language models (LLMs) have demonstrated strong potential in detecting fraudulent behavior by analyzing transcribed phone conversations. However, these capabilities introduce notable privacy risks, as such conversations frequently contain sensitive personal information that may be exposed to third-party service providers during processing. In this work, we explore how to harness LLMs for phone scam detection while preserving user privacy. We propose MASK (Modular Adaptive Sanitization Kit), a trainable and extensible framework that enables dynamic privacy adjustment based on individual preferences. MASK provides a pluggable architecture that accommodates diverse sanitization methods - from traditional keyword-based techniques for high-privacy users to sophisticated neural approaches for those prioritizing accuracy. We also discuss potential modeling approaches and loss function designs for future development, enabling the creation of truly personalized, privacy-aware LLM-based detection systems that balance user trust and detection effectiveness, even beyond phone scam context.
Authors: Wei-Chia Chang, Yan-Ann Chen
Abstract: Vehicle make and model recognition (VMMR) is an important task in intelligent transportation systems, but existing approaches struggle to adapt to newly released models. Contrastive Language-Image Pretraining (CLIP) provides strong visual-text alignment, yet its fixed pretrained weights limit performance without costly image-specific finetuning. We propose a pipeline that integrates vision language models (VLMs) with Retrieval-Augmented Generation (RAG) to support zero-shot recognition through text-based reasoning. A VLM converts vehicle images into descriptive attributes, which are compared against a database of textual features. Relevant entries are retrieved and combined with the description to form a prompt, and a language model (LM) infers the make and model. This design avoids large-scale retraining and enables rapid updates by adding textual descriptions of new vehicles. Experiments show that the proposed method improves recognition by nearly 20% over the CLIP baseline, demonstrating the potential of RAG-enhanced LM reasoning for scalable VMMR in smart-city applications.
Authors: Giovanni De Muri, Mark Vero, Robin Staab, Martin Vechev
Abstract: LLMs are often used by downstream users as teacher models for knowledge distillation, compressing their capabilities into memory-efficient models. However, as these teacher models may stem from untrusted parties, distillation can raise unexpected security risks. In this paper, we investigate the security implications of knowledge distillation from backdoored teacher models. First, we show that prior backdoors mostly do not transfer onto student models. Our key insight is that this is because existing LLM backdooring methods choose trigger tokens that rarely occur in usual contexts. We argue that this underestimates the security risks of knowledge distillation and introduce a new backdooring technique, T-MTB, that enables the construction and study of transferable backdoors. T-MTB carefully constructs a composite backdoor trigger, made up of several specific tokens that often occur individually in anticipated distillation datasets. As such, the poisoned teacher remains stealthy, while during distillation the individual presence of these tokens provides enough signal for the backdoor to transfer onto the student. Using T-MTB, we demonstrate and extensively study the security risks of transferable backdoors across two attack scenarios, jailbreaking and content modulation, and across four model families of LLMs.
Authors: Zebin Yang, Sunjian Zheng, Tong Xie, Tianshi Xu, Bo Yu, Fan Wang, Jie Tang, Shaoshan Liu, Meng Li
Abstract: Object-goal navigation (ObjNav) tasks an agent with navigating to the location of a specific object in an unseen environment. Embodied agents equipped with large language models (LLMs) and online constructed navigation maps can perform ObjNav in a zero-shot manner. However, existing agents heavily rely on giant LLMs on the cloud, e.g., GPT-4, while directly switching to small LLMs, e.g., LLaMA3.2-11b, suffer from significant success rate drops due to limited model capacity for understanding complex navigation maps, which prevents deploying ObjNav on local devices. At the same time, the long prompt introduced by the navigation map description will cause high planning latency on local devices. In this paper, we propose EfficientNav to enable on-device efficient LLM-based zero-shot ObjNav. To help the smaller LLMs better understand the environment, we propose semantics-aware memory retrieval to prune redundant information in navigation maps. To reduce planning latency, we propose discrete memory caching and attention-based memory clustering to efficiently save and re-use the KV cache. Extensive experimental results demonstrate that EfficientNav achieves 11.1% improvement in success rate on HM3D benchmark over GPT-4-based baselines, and demonstrates 6.7x real-time latency reduction and 4.7x end-to-end latency reduction over GPT-4 planner. Our code will be released soon.
Authors: Loc Phuc Truong Nguyen, Hung Thanh Do
Abstract: As AI systems enter high-stakes domains, evaluation must extend beyond predictive accuracy to include explainability, fairness, robustness, and sustainability. We introduce RAISE (Responsible AI Scoring and Evaluation), a unified framework that quantifies model performance across these four dimensions and aggregates them into a single, holistic Responsibility Score. We evaluated three deep learning models: a Multilayer Perceptron (MLP), a Tabular ResNet, and a Feature Tokenizer Transformer, on structured datasets from finance, healthcare, and socioeconomics. Our findings reveal critical trade-offs: the MLP demonstrated strong sustainability and robustness, the Transformer excelled in explainability and fairness at a very high environmental cost, and the Tabular ResNet offered a balanced profile. These results underscore that no single model dominates across all responsibility criteria, highlighting the necessity of multi-dimensional evaluation for responsible model selection. Our implementation is available at: https://github.com/raise-framework/raise.
Authors: Chunyang Li, Yilun Zheng, Xinting Huang, Tianqing Fang, Jiahao Xu, Yangqiu Song, Lihui Chen, Han Hu
Abstract: The paradigm of LLM-as-a-judge is emerging as a scalable and efficient alternative to human evaluation, demonstrating strong performance on well-defined tasks. However, its reliability in open-ended tasks with dynamic environments and complex interactions remains unexplored. To bridge the gap, we introduce WebDevJudge, a systematic benchmark for assessing LLM-as-a-judge performance in web development, with support for both non-interactive evaluation based on static observations and continuous interactive evaluation with a dynamic web environment. WebDevJudge comprises human preference labels over paired web implementations, annotated with structured and query-grounded rubrics to ensure high-quality ground truth. Using this benchmark, we comprehensively evaluate various evaluators, including LLMs, MLLMs, and agentic workflows. We systematically investigate the impact of different paradigms and guidance mechanisms. Our experiments reveal a significant gap between LLM judges and human experts. In-depth analysis indicates this gap stems from fundamental model limitations, including failures in recognizing functional equivalence, verifying task feasibility, and mitigating bias. Overall, WebDevJudge presents a significant challenge to LLM-as-a-judge, offering insights to guide future research toward developing more reliable and capable automated evaluators for complicated scenarios. Code and data are available at https://github.com/lcy2723/WebDevJudge.
Authors: Tja\v{s}a Ar\v{c}on, Marko Robnik-\v{S}ikonja, Polona Tratnik
Abstract: Artificial intelligence approaches are being adapted to many research areas, including digital humanities. We built a methodology for large-scale analyses in folkloristics. Using machine learning and natural language processing, we automatically detected motifs in a large collection of Cinderella variants and analysed their similarities and differences with clustering and dimensionality reduction. The results show that large language models detect complex interactions in tales, enabling computational analysis of extensive text collections and facilitating cross-lingual comparisons.
Authors: Zhenxing Zhang, Jiayan Teng, Zhuoyi Yang, Tiankun Cao, Cheng Wang, Xiaotao Gu, Jie Tang, Dan Guo, Meng Wang
Abstract: We present Kaleido, a subject-to-video~(S2V) generation framework, which aims to synthesize subject-consistent videos conditioned on multiple reference images of target subjects. Despite recent progress in S2V generation models, existing approaches remain inadequate at maintaining multi-subject consistency and at handling background disentanglement, often resulting in lower reference fidelity and semantic drift under multi-image conditioning. These shortcomings can be attributed to several factors. Primarily, the training dataset suffers from a lack of diversity and high-quality samples, as well as cross-paired data, i.e., paired samples whose components originate from different instances. In addition, the current mechanism for integrating multiple reference images is suboptimal, potentially resulting in the confusion of multiple subjects. To overcome these limitations, we propose a dedicated data construction pipeline, incorporating low-quality sample filtering and diverse data synthesis, to produce consistency-preserving training data. Moreover, we introduce Reference Rotary Positional Encoding (R-RoPE) to process reference images, enabling stable and precise multi-image integration. Extensive experiments across numerous benchmarks demonstrate that Kaleido significantly outperforms previous methods in consistency, fidelity, and generalization, marking an advance in S2V generation.
Authors: Katerina Drakos, Eva Paraschou, Simay Toplu, Line Harder Clemmensen, Christoph L\"utge, Nicole Nadine L{\o}nfeldt, Sneha Das
Abstract: Artificial intelligence has been introduced as a way to improve access to mental health support. However, most AI mental health chatbots rely on a limited range of disciplinary input, and fail to integrate expertise across the chatbot's lifecycle. This paper examines the cost-benefit trade-off of interdisciplinary collaboration in AI mental health chatbots. We argue that involving experts from technology, healthcare, ethics, and law across key lifecycle phases is essential to ensure value-alignment and compliance with the high-risk requirements of the AI Act. We also highlight practical recommendations and existing frameworks to help balance the challenges and benefits of interdisciplinarity in mental health chatbots.
Authors: Gilles Audemard, Sylvie Coste-Marquis, Pierre Marquis, Mehdi Sabiri, Nicolas Szczepanski
Abstract: We present a new approach for distilling boosted trees into decision trees, in the objective of generating an ML model offering an acceptable compromise in terms of predictive performance and interpretability. We explain how the correction approach called rectification can be used to implement such a distillation process. We show empirically that this approach provides interesting results, in comparison with an approach to distillation achieved by retraining the model.
Authors: Zhangquan Chen, Manyuan Zhang, Xinlei Yu, Xufang Luo, Mingze Sun, Zihao Pan, Yan Feng, Peng Pei, Xunliang Cai, Ruqi Huang
Abstract: Though recent advances in vision-language models (VLMs) have achieved remarkable progress across a wide range of multimodal tasks, understanding 3D spatial relationships from limited views remains a significant challenge. Previous reasoning methods typically rely on pure text (e.g., topological cognitive maps) or on 2D visual cues. However, their limited representational capacity hinders performance in specific tasks that require 3D spatial imagination. To address this limitation, we propose 3DThinker, a framework that can effectively exploits the rich geometric information embedded within images while reasoning, like humans do. Our framework is the first to enable 3D mentaling during reasoning without any 3D prior input, and it does not rely on explicitly labeled 3D data for training. Specifically, our training consists of two stages. First, we perform supervised training to align the 3D latent generated by VLM while reasoning with that of a 3D foundation model (e.g., VGGT). Then, we optimize the entire reasoning trajectory solely based on outcome signals, thereby refining the underlying 3D mentaling. Extensive experiments across multiple benchmarks show that 3DThinker consistently outperforms strong baselines and offers a new perspective toward unifying 3D representations into multimodal reasoning. Our code will be available at https://github.com/zhangquanchen/3DThinker.
Authors: Baptiste Bauvin, Lo\"ic Baret, Ola Ahmad
Abstract: Neural network compression has gained increasing attention in recent years, particularly in computer vision applications, where the need for model reduction is crucial for overcoming deployment constraints. Pruning is a widely used technique that prompts sparsity in model structures, e.g. weights, neurons, and layers, reducing size and inference costs. Structured pruning is especially important as it allows for the removal of entire structures, which further accelerates inference time and reduces memory overhead. However, it can be computationally expensive, requiring iterative retraining and optimization. To overcome this problem, recent methods considered one-shot setting, which applies pruning directly at post-training. Unfortunately, they often lead to a considerable drop in performance. In this paper, we focus on this issue by proposing a novel one-shot pruning framework that relies on explainable deep learning. First, we introduce a causal-aware pruning approach that leverages cause-effect relations between model predictions and structures in a progressive pruning process. It allows us to efficiently reduce the size of the network, ensuring that the removed structures do not deter the performance of the model. Then, through experiments conducted on convolution neural network and vision transformer baselines, pre-trained on classification tasks, we demonstrate that our method consistently achieves substantial reductions in model size, with minimal impact on performance, and without the need for fine-tuning. Overall, our approach outperforms its counterparts, offering the best trade-off. Our code is available on GitHub.
Authors: Sheida Rahnamai Kordasiabi, Damian Dalle Nogare, Florian Jug
Abstract: Semantic segmentation of electron microscopy (EM) images of biological samples remains a challenge in the life sciences. EM data captures details of biological structures, sometimes with such complexity that even human observers can find it overwhelming. We introduce {\epsilon}-Seg, a method based on hierarchical variational autoencoders (HVAEs), employing center-region masking, sparse label contrastive learning (CL), a Gaussian mixture model (GMM) prior, and clustering-free label prediction. Center-region masking and the inpainting loss encourage the model to learn robust and representative embeddings to distinguish the desired classes, even if training labels are sparse (0.05% of the total image data or less). For optimal performance, we employ CL and a GMM prior to shape the latent space of the HVAE such that encoded input patches tend to cluster wrt. the semantic classes we wish to distinguish. Finally, instead of clustering latent embeddings for semantic segmentation, we propose a MLP semantic segmentation head to directly predict class labels from latent embeddings. We show empirical results of {\epsilon}-Seg and baseline methods on 2 dense EM datasets of biological tissues and demonstrate the applicability of our method also on fluorescence microscopy data. Our results show that {\epsilon}-Seg is capable of achieving competitive sparsely-supervised segmentation results on complex biological image data, even if only limited amounts of training labels are available.
Authors: Kyo Kuroki, Yasuyuki Okoshi, Thiem Van Chu, Kazushi Kawamura, Masato Motomura
Abstract: This paper proposes a novel matrix quantization method, Binary Quadratic Quantization (BQQ). In contrast to conventional first-order quantization approaches, such as uniform quantization and binary coding quantization, that approximate real-valued matrices via linear combinations of binary bases, BQQ leverages the expressive power of binary quadratic expressions while maintaining an extremely compact data format. We validate our approach with two experiments: a matrix compression benchmark and post-training quantization (PTQ) on pretrained Vision Transformer-based models. Experimental results demonstrate that BQQ consistently achieves a superior trade-off between memory efficiency and reconstruction error than conventional methods for compressing diverse matrix data. It also delivers strong PTQ performance, even though we neither target state-of-the-art PTQ accuracy under tight memory constraints nor rely on PTQ-specific binary matrix optimization. For example, our proposed method outperforms the state-of-the-art PTQ method by up to 2.2\% and 59.1% on the ImageNet dataset under the calibration-based and data-free scenarios, respectively, with quantization equivalent to 2 bits. These findings highlight the surprising effectiveness of binary quadratic expressions for efficient matrix approximation and neural network compression.
Authors: Qi Li, Junpan Wu, Xiang Liu, Yuxin Wang, Zeyu Li, Zhenheng Tang, Yuhan Chen, Shaohuai Shi, Xiaowen Chu
Abstract: The reasoning large language model (RLLM) has been proven competitive in solving complex reasoning tasks such as mathematics, coding, compared to general LLM. However, the serving performance and behavior of RLLM remains unexplored, which may undermine the deployment and utilization of RLLM in real-world scenario. To close this gap, in this paper, we conduct a comprehensive study of RLLM service. We first perform a pilot study on comparing the serving performance between RLLM and traditional LLM and reveal that there are several distinct differences regarding serving behavior: (1) significant memory usage and fluctuations; (2) straggler requests; (3) adaptive running time; (4) domain preference. Then we further investigate whether existing inference optimization techniques are valid for RLLM. Our main takeaways are that model quantization methods and speculative decoding can improve service system efficiency with small compromise to RLLM accuracy, while prefix caching, KV cache quantization may even degrade accuracy or serving performance for small RLLM. Lastly, we conduct evaluation under real world workload modeled by Gamma distribution to verify our findings. Empirical results of real world workload evaluation across different dataset are aligned with our main findings regarding RLLM serving. We hope our work can provide the research community and industry with insights to advance RLLM inference serving.
Authors: Alexander Nemecek, Zebin Yun, Zahra Rahmani, Yaniv Harel, Vipin Chaudhary, Mahmood Sharif, Erman Ayday
Abstract: As large language models (LLMs) become progressively more embedded in clinical decision-support, documentation, and patient-information systems, ensuring their privacy and trustworthiness has emerged as an imperative challenge for the healthcare sector. Fine-tuning LLMs on sensitive electronic health record (EHR) data improves domain alignment but also raises the risk of exposing patient information through model behaviors. In this work-in-progress, we present an exploratory empirical study on membership inference vulnerabilities in clinical LLMs, focusing on whether adversaries can infer if specific patient records were used during model training. Using a state-of-the-art clinical question-answering model, Llemr, we evaluate both canonical loss-based attacks and a domain-motivated paraphrasing-based perturbation strategy that more realistically reflects clinical adversarial conditions. Our preliminary findings reveal limited but measurable membership leakage, suggesting that current clinical LLMs provide partial resistance yet remain susceptible to subtle privacy risks that could undermine trust in clinical AI adoption. These results motivate continued development of context-aware, domain-specific privacy evaluations and defenses such as differential privacy fine-tuning and paraphrase-aware training, to strengthen the security and trustworthiness of healthcare AI systems.
Authors: Michael J. Wooldridge, Attila Bagoly, Jonathan J. Ward, Emanuele La Malfa, Gabriel Paludo Licks
Abstract: Recent surges in LLM-driven intelligent systems largely overlook decades of foundational multi-agent systems (MAS) research, resulting in frameworks with critical limitations such as centralization and inadequate trust and communication protocols. This paper introduces the Fetch.ai architecture, an industrial-strength platform designed to bridge this gap by facilitating the integration of classical MAS principles with modern AI capabilities. We present a novel, multi-layered solution built on a decentralized foundation of on-chain blockchain services for verifiable identity, discovery, and transactions. This is complemented by a comprehensive development framework for creating secure, interoperable agents, a cloud-based platform for deployment, and an intelligent orchestration layer where an agent-native LLM translates high-level human goals into complex, multi-agent workflows. We demonstrate the deployed nature of this system through a decentralized logistics use case where autonomous agents dynamically discover, negotiate, and transact with one another securely. Ultimately, the Fetch.ai stack provides a principled architecture for moving beyond current agent implementations towards open, collaborative, and economically sustainable multi-agent ecosystems.
Authors: Joongkyu Lee, Seouh-won Yi, Min-hwan Oh
Abstract: We study online preference-based reinforcement learning (PbRL) with the goal of improving sample efficiency. While a growing body of theoretical work has emerged-motivated by PbRL's recent empirical success, particularly in aligning large language models (LLMs)-most existing studies focus only on pairwise comparisons. A few recent works (Zhu et al., 2023, Mukherjee et al., 2024, Thekumparampil et al., 2024) have explored using multiple comparisons and ranking feedback, but their performance guarantees fail to improve-and can even deteriorate-as the feedback length increases, despite the richer information available. To address this gap, we adopt the Plackett-Luce (PL) model for ranking feedback over action subsets and propose M-AUPO, an algorithm that selects multiple actions by maximizing the average uncertainty within the offered subset. We prove that M-AUPO achieves a suboptimality gap of $\tilde{\mathcal{O}}\left( \frac{d}{T} \sqrt{ \sum_{t=1}^T \frac{1}{|S_t|}} \right)$, where $T$ is the total number of rounds, $d$ is the feature dimension, and $|S_t|$ is the size of the subset at round $t$. This result shows that larger subsets directly lead to improved performance and, notably, the bound avoids the exponential dependence on the unknown parameter's norm, which was a fundamental limitation in most previous works. Moreover, we establish a near-matching lower bound of $\Omega \left( \frac{d}{K \sqrt{T}} \right)$, where $K$ is the maximum subset size. To the best of our knowledge, this is the first theoretical result in PbRL with ranking feedback that explicitly shows improved sample efficiency as a function of the subset size.
Authors: Chengwen Du, Tao Chen
Abstract: To mitigate unfair and unethical discrimination over sensitive features (e.g., gender, age, or race), fairness testing plays an integral role in engineering systems that leverage AI models to handle tabular data. A key challenge therein is how to effectively reveal fairness bugs under an intractable sample size using perturbation. Much current work has been focusing on designing the test sample generators, ignoring the valuable knowledge about data characteristics that can help guide the perturbation and hence limiting their full potential. In this paper, we seek to bridge such a gap by proposing a generic framework of causally perturbed fairness testing, dubbed CausalFT. Through causal inference, the key idea of CausalFT is to extract the most directly and causally relevant non-sensitive feature to its sensitive counterpart, which can jointly influence the prediction of the label. Such a causal relationship is then seamlessly injected into the perturbation to guide a test sample generator. Unlike existing generator-level work, CausalFT serves as a higher-level framework that can be paired with diverse base generators. Extensive experiments on 1296 cases confirm that CausalFT can considerably improve arbitrary base generators in revealing fairness bugs over 93% of the cases with acceptable extra runtime overhead. Compared with a state-of-the-art approach that ranks the non-sensitive features solely based on correlation, CausalFT performs significantly better on 64% cases while being much more efficient. Further, CausalFT can better improve bias resilience in nearly all cases.
Authors: Sidhant Narula, Javad Rafiei Asl, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi
Abstract: Large Language Models (LLMs) remain vulnerable to multi-turn jailbreak attacks. We introduce HarmNet, a modular framework comprising ThoughtNet, a hierarchical semantic network; a feedback-driven Simulator for iterative query refinement; and a Network Traverser for real-time adaptive attack execution. HarmNet systematically explores and refines the adversarial space to uncover stealthy, high-success attack paths. Experiments across closed-source and open-source LLMs show that HarmNet outperforms state-of-the-art methods, achieving higher attack success rates. For example, on Mistral-7B, HarmNet achieves a 99.4% attack success rate, 13.9% higher than the best baseline. Index terms: jailbreak attacks; large language models; adversarial framework; query refinement.
Authors: Ming Li
Abstract: Large Language Models demonstrate strong capabilities in single-turn instruction following but suffer from Lost-in-Conversation (LiC), a degradation in performance as information is revealed progressively in multi-turn settings. Motivated by the current progress on Reinforcement Learning with Verifiable Rewards (RLVR), we propose Curriculum Reinforcement Learning with Verifiable Accuracy and Abstention Rewards (RLAAR), a framework that encourages models not only to generate correct answers, but also to judge the solvability of questions in the multi-turn conversation setting. Our approach employs a competence-gated curriculum that incrementally increases dialogue difficulty (in terms of instruction shards), stabilizing training while promoting reliability. Using multi-turn, on-policy rollouts and a mixed-reward system, RLAAR teaches models to balance problem-solving with informed abstention, reducing premature answering behaviors that cause LiC. Evaluated on LiC benchmarks, RLAAR significantly mitigates LiC performance decay (62.6% to 75.1%) and improves calibrated abstention rates (33.5% to 73.4%). Together, these results provide a practical recipe for building multi-turn reliable and trustworthy LLMs.
Authors: Vik Pant, Eric Yu
Abstract: Modern socio-technical systems are characterized by strategic coopetition where actors simultaneously cooperate to create value and compete to capture it. While conceptual modeling languages like i* provide rich qualitative representations of strategic dependencies, they lack mechanisms for quantitative analysis of dynamic trade-offs. Conversely, classical game theory offers mathematical rigor but strips away contextual richness. This technical report bridges this gap by developing computational foundations that formalize two critical dimensions of coopetition: interdependence and complementarity. We ground interdependence in i* structural dependency analysis, translating depender-dependee-dependum relationships into quantitative interdependence coefficients through a structured translation framework. We formalize complementarity following Brandenburger and Nalebuff's Added Value concept, modeling synergistic value creation with validated parameterization. We integrate structural dependencies with bargaining power in value appropriation and introduce a game-theoretic formulation where Nash Equilibrium incorporates structural interdependence. Validation combines comprehensive experimental testing across power and logarithmic value function specifications, demonstrating functional form robustness, with empirical application to the Samsung-Sony S-LCD joint venture (2004-2011), where logarithmic specifications achieve superior empirical fit (validation score 45/60) while power functions provide theoretical tractability. This technical report serves as the foundational reference for a coordinated research program examining strategic coopetition in requirements engineering and multi-agent systems, with companion work addressing trust dynamics, team production, and reciprocity mechanisms.
Authors: Mengqi Li, Lei Zhao, Anthony Man-Cho So, Ruoyu Sun, Xiao Li
Abstract: We present a simple, self-help online supervised finetuning (OSFT) paradigm for LLM reasoning. In this paradigm, the model generates its own responses and is immediately finetuned on this self-generated data. OSFT is a highly efficient training strategy for LLM reasoning, as it is reward-free and uses just one rollout by default. Experiment results show that OSFT achieves downstream performance on challenging mathematical reasoning tasks comparable to strong reinforcement learning with verifiable rewards (RLVR) methods such as GRPO. Our ablation study further demonstrates the efficiency and robustness of OSFT. The major mechanism of OSFT lies in facilitating the model's own existing preference (latent knowledge) learned from pretraining, which leads to reasoning ability improvement. We believe that OSFT offers an efficient and promising alternative to more complex, reward-based training paradigms. Our code is available at https://github.com/ElementQi/OnlineSFT.
Authors: Shuxin Lin, Dhaval Patel, Christodoulos Constantinides
Abstract: Small Language Models (SLMs) are becoming increasingly popular in specialized fields, such as industrial applications, due to their efficiency, lower computational requirements, and ability to be fine-tuned for domain-specific tasks, enabling accurate and cost-effective solutions. However, performing complex reasoning using SLMs in specialized fields such as Industry 4.0 remains challenging. In this paper, we propose a knowledge distillation framework for industrial asset health, which transfers reasoning capabilities via Chain-of-Thought (CoT) distillation from Large Language Models (LLMs) to smaller, more efficient models (SLMs). We discuss the advantages and the process of distilling LLMs using multi-choice question answering (MCQA) prompts to enhance reasoning and refine decision-making. We also perform in-context learning to verify the quality of the generated knowledge and benchmark the performance of fine-tuned SLMs with generated knowledge against widely used LLMs. The results show that the fine-tuned SLMs with CoT reasoning outperform the base models by a significant margin, narrowing the gap to their LLM counterparts. Our code is open-sourced at: https://github.com/IBM/FailureSensorIQ.
Authors: Neel Patel, Alexander Wong, Ashkan Ebadi
Abstract: Tuberculosis remains a critical global health issue, particularly in resource-limited and remote areas. Early detection is vital for treatment, yet the lack of skilled radiologists underscores the need for artificial intelligence (AI)-driven screening tools. Developing reliable AI models is challenging due to the necessity for large, high-quality datasets, which are costly to obtain. To tackle this, we propose a teacher--student framework which enhances both disease and symptom detection on chest X-rays by integrating two supervised heads and a self-supervised head. Our model achieves an accuracy of 98.85% for distinguishing between COVID-19, tuberculosis, and normal cases, and a macro-F1 score of 90.09% for multilabel symptom detection, significantly outperforming baselines. The explainability assessments also show the model bases its predictions on relevant anatomical features, demonstrating promise for deployment in clinical screening and triage settings.
Authors: Yigit Korkmaz, Urvi Bhuwania, Ayush Jain, Erdem B{\i}y{\i}k
Abstract: Value-based algorithms are a cornerstone of off-policy reinforcement learning due to their simplicity and training stability. However, their use has traditionally been restricted to discrete action spaces, as they rely on estimating Q-values for individual state-action pairs. In continuous action spaces, evaluating the Q-value over the entire action space becomes computationally infeasible. To address this, actor-critic methods are typically employed, where a critic is trained on off-policy data to estimate Q-values, and an actor is trained to maximize the critic's output. Despite their popularity, these methods often suffer from instability during training. In this work, we propose a purely value-based framework for continuous control that revisits structural maximization of Q-functions, introducing a set of key architectural and algorithmic choices to enable efficient and stable learning. We evaluate the proposed actor-free Q-learning approach on a range of standard simulation tasks, demonstrating performance and sample efficiency on par with state-of-the-art baselines, without the cost of learning a separate actor. Particularly, in environments with constrained action spaces, where the value functions are typically non-smooth, our method with structural maximization outperforms traditional actor-critic methods with gradient-based maximization. We have released our code at https://github.com/USC-Lira/Q3C.
Authors: Chenghao Zhu, Meiling Tao, Tiannan Wang, Dongyi Ding, Yuchen Eleanor Jiang, Wangchunshu Zhou
Abstract: Faithfully personalizing large language models (LLMs) to align with individual user preferences is a critical but challenging task. While supervised fine-tuning (SFT) quickly reaches a performance plateau, standard reinforcement learning from human feedback (RLHF) also struggles with the nuances of personalization. Scalar-based reward models are prone to reward hacking which leads to verbose and superficially personalized responses. To address these limitations, we propose Critique-Post-Edit, a robust reinforcement learning framework that enables more faithful and controllable personalization. Our framework integrates two key components: (1) a Personalized Generative Reward Model (GRM) that provides multi-dimensional scores and textual critiques to resist reward hacking, and (2) a Critique-Post-Edit mechanism where the policy model revises its own outputs based on these critiques for more targeted and efficient learning. Under a rigorous length-controlled evaluation, our method substantially outperforms standard PPO on personalization benchmarks. Personalized Qwen2.5-7B achieves an average 11\% win-rate improvement, and personalized Qwen2.5-14B model surpasses the performance of GPT-4.1. These results demonstrate a practical path to faithful, efficient, and controllable personalization.
Authors: Rongyuan Wu, Lingchen Sun, Zhengqiang Zhang, Shihao Wang, Tianhe Wu, Qiaosi Yi, Shuai Li, Lei Zhang
Abstract: Benefiting from pre-trained text-to-image (T2I) diffusion models, real-world image super-resolution (Real-ISR) methods can synthesize rich and realistic details. However, due to the inherent stochasticity of T2I models, different noise inputs often lead to outputs with varying perceptual quality. Although this randomness is sometimes seen as a limitation, it also introduces a wider perceptual quality range, which can be exploited to improve Real-ISR performance. To this end, we introduce Direct Perceptual Preference Optimization for Real-ISR (DP$^2$O-SR), a framework that aligns generative models with perceptual preferences without requiring costly human annotations. We construct a hybrid reward signal by combining full-reference and no-reference image quality assessment (IQA) models trained on large-scale human preference datasets. This reward encourages both structural fidelity and natural appearance. To better utilize perceptual diversity, we move beyond the standard best-vs-worst selection and construct multiple preference pairs from outputs of the same model. Our analysis reveals that the optimal selection ratio depends on model capacity: smaller models benefit from broader coverage, while larger models respond better to stronger contrast in supervision. Furthermore, we propose hierarchical preference optimization, which adaptively weights training pairs based on intra-group reward gaps and inter-group diversity, enabling more efficient and stable learning. Extensive experiments across both diffusion- and flow-based T2I backbones demonstrate that DP$^2$O-SR significantly improves perceptual quality and generalizes well to real-world benchmarks.
Authors: Nutkritta Kraipatthanapong, Natthaphat Thathong, Pannita Suksawas, Thanunnut Klunklin, Kritin Vongthonglua, Krit Attahakul, Aueaphum Aueawatthanaphisut
Abstract: This paper presents a novel Lyapunov-Based Quantum Reinforcement Learning (LQRL) framework that integrates quantum policy optimization with Lyapunov stability analysis for continuous-time vehicle control. The proposed approach combines the representational power of variational quantum circuits (VQCs) with a stability-aware policy gradient mechanism to ensure asymptotic convergence and safe decision-making under dynamic environments. The vehicle longitudinal control problem was formulated as a continuous-state reinforcement learning task, where the quantum policy network generates control actions subject to Lyapunov stability constraints. Simulation experiments were conducted in a closed-loop adaptive cruise control scenario using a quantum-inspired policy trained under stability feedback. The results demonstrate that the LQRL framework successfully embeds Lyapunov stability verification into quantum policy learning, enabling interpretable and stability-aware control performance. Although transient overshoot and Lyapunov divergence were observed under aggressive acceleration, the system maintained bounded state evolution, validating the feasibility of integrating safety guarantees within quantum reinforcement learning architectures. The proposed framework provides a foundational step toward provably safe quantum control in autonomous systems and hybrid quantum-classical optimization domains.
Authors: Ling Team, Anqi Shen, Baihui Li, Bin Hu, Bin Jing, Cai Chen, Chao Huang, Chao Zhang, Chaokun Yang, Cheng Lin, Chengyao Wen, Congqi Li, Deng Zhao, Dingbo Yuan, Donghai You, Fagui Mao, Fanzhuang Meng, Feng Xu, Guojie Li, Guowei Wang, Hao Dai, Haonan Zheng, Hong Liu, Jia Guo, Jiaming Liu, Jian Liu, Jianhao Fu, Jiannan Shi, Jianwen Wang, Jianxin Lai, Jin Yang, Jun Mei, Jun Zhou, Junbo Zhao, Junping Zhao, Kuan Xu, Le Su, Lei Chen, Li Tang, Liang Jiang, Liangcheng Fu, Lianhao Xu, Linfeng Shi, Lisha Liao, Longfei Zheng, Meng Li, Mingchun Chen, Qi Zuo, Qiang Cheng, Qianggang Cao, Qitao Shi, Quanrui Guo, Senlin Zhu, Shaofei Wang, Shaomian Zheng, Shuaicheng Li, Shuwei Gu, Siba Chen, Tao Wu, Tao Zhang, Tianyu Zhang, Tianyu Zhou, Tiwei Bie, Tongkai Yang, Wang Hong, Wang Ren, Weihua Chen, Wenbo Yu, Wengang Zheng, Xiangchun Wang, Xiaodong Yan, Xiaopei Wan, Xin Zhao, Xinyu Kong, Xinyu Tang, Xudong Han, Xudong Wang, Xuemin Yang, Xueyu Hu, Yalin Zhang, Yan Sun, Yicheng Shan, Yilong Wang, Yingying Xu, Yongkang Liu, Yongzhen Guo, Yuanyuan Wang, Yuchen Yan, Yuefan Wang, Yuhong Guo, Zehuan Li, Zhankai Xu, Zhe Li, Zhenduo Zhang, Zhengke Gui, Zhenxuan Pan, Zhenyu Huang, Zhenzhong Lan, Zhiqiang Ding, Zhiqiang Zhang, Zhixun Li, Zhizhen Liu, Zihao Wang, Zujie Wen
Abstract: We present Ring-1T, the first open-source, state-of-the-art thinking model with a trillion-scale parameter. It features 1 trillion total parameters and activates approximately 50 billion per token. Training such models at a trillion-parameter scale introduces unprecedented challenges, including train-inference misalignment, inefficiencies in rollout processing, and bottlenecks in the RL system. To address these, we pioneer three interconnected innovations: (1) IcePop stabilizes RL training via token-level discrepancy masking and clipping, resolving instability from training-inference mismatches; (2) C3PO++ improves resource utilization for long rollouts under a token budget by dynamically partitioning them, thereby obtaining high time efficiency; and (3) ASystem, a high-performance RL framework designed to overcome the systemic bottlenecks that impede trillion-parameter model training. Ring-1T delivers breakthrough results across critical benchmarks: 93.4 on AIME-2025, 86.72 on HMMT-2025, 2088 on CodeForces, and 55.94 on ARC-AGI-v1. Notably, it attains a silver medal-level result on the IMO-2025, underscoring its exceptional reasoning capabilities. By releasing the complete 1T parameter MoE model to the community, we provide the research community with direct access to cutting-edge reasoning capabilities. This contribution marks a significant milestone in democratizing large-scale reasoning intelligence and establishes a new baseline for open-source model performance.
Authors: Jizhan Fang, Xinle Deng, Haoming Xu, Ziyan Jiang, Yuqi Tang, Ziwen Xu, Shumin Deng, Yunzhi Yao, Mengru Wang, Shuofei Qiao, Huajun Chen, Ningyu Zhang
Abstract: Despite their remarkable capabilities, Large Language Models (LLMs) struggle to effectively leverage historical interaction information in dynamic and complex environments. Memory systems enable LLMs to move beyond stateless interactions by introducing persistent information storage, retrieval, and utilization mechanisms. However, existing memory systems often introduce substantial time and computational overhead. To this end, we introduce a new memory system called LightMem, which strikes a balance between the performance and efficiency of memory systems. Inspired by the Atkinson-Shiffrin model of human memory, LightMem organizes memory into three complementary stages. First, cognition-inspired sensory memory rapidly filters irrelevant information through lightweight compression and groups information according to their topics. Next, topic-aware short-term memory consolidates these topic-based groups, organizing and summarizing content for more structured access. Finally, long-term memory with sleep-time update employs an offline procedure that decouples consolidation from online inference. Experiments on LongMemEval with GPT and Qwen backbones show that LightMem outperforms strong baselines in accuracy (up to 10.9% gains) while reducing token usage by up to 117x, API calls by up to 159x, and runtime by over 12x. The code is available at https://github.com/zjunlp/LightMem.
Authors: Akshat Gupta, Jay Yeung, Gopala Anumanchipalli, Anna Ivanova
Abstract: Growing evidence suggests that large language models do not use their depth uniformly, yet we still lack a fine-grained understanding of their layer-wise prediction dynamics. In this paper, we trace the intermediate representations of several open-weight models during inference and reveal a structured and nuanced use of depth. Specifically, we propose a "Guess-then-Refine" framework that explains how LLMs internally structure their computations to make predictions. We first show that the top-ranked predictions in early LLM layers are composed primarily of high-frequency tokens, which act as statistical guesses proposed by the model early on due to the lack of appropriate contextual information. As contextual information develops deeper into the model, these initial guesses get refined into contextually appropriate tokens. Even high-frequency token predictions from early layers get refined >70% of the time, indicating that correct token prediction is not "one-and-done". We then go beyond frequency-based prediction to examine the dynamic usage of layer depth across three case studies. (i) Part-of-speech analysis shows that function words are, on average, the earliest to be predicted correctly. (ii) Fact recall task analysis shows that, in a multi-token answer, the first token requires more computational depth than the rest. (iii) Multiple-choice task analysis shows that the model identifies the format of the response within the first half of the layers, but finalizes its response only toward the end. Together, our results provide a detailed view of depth usage in LLMs, shedding light on the layer-by-layer computations that underlie successful predictions and providing insights for future works to improve computational efficiency in transformer-based models.
Authors: Haochen Wang, Yuhao Wang, Tao Zhang, Yikang Zhou, Yanwei Li, Jiacong Wang, Ye Tian, Jiahao Meng, Zilong Huang, Guangcan Mai, Anran Wang, Yunhai Tong, Zhuochen Wang, Xiangtai Li, Zhaoxiang Zhang
Abstract: While Multimodal Large Language Models (MLLMs) excel at holistic understanding, they struggle in capturing the dense world with complex scenes, requiring fine-grained analysis of intricate details and object inter-relationships. Region-level MLLMs have been a promising step. However, previous attempts are generally optimized to understand given regions in isolation, neglecting crucial global contexts. To address this, we introduce Grasp Any Region (GAR) for comprehen- sive region-level visual understanding. Empowered by an effective RoI-aligned feature replay technique, GAR supports (1) precise perception by leveraging necessary global contexts, and (2) modeling interactions between multiple prompts. Together, it then naturally achieves (3) advanced compositional reasoning to answer specific free-form questions about any region, shifting the paradigm from passive description to active dialogue. Moreover, we construct GAR-Bench, which not only provides a more accurate evaluation of single-region comprehension, but also, more importantly, measures interactions and complex reasoning across multiple regions. Extensive experiments have demonstrated that GAR-1B not only maintains the state-of-the-art captioning capabilities, e.g., outperforming DAM-3B +4.5 on DLC-Bench, but also excels at modeling relationships between multiple prompts with advanced comprehension capabilities, even surpassing InternVL3-78B on GAR-Bench-VQA. More importantly, our zero-shot GAR-8B even outperforms in-domain VideoRefer-7B on VideoRefer-BenchQ, indicating its strong capabilities can be easily transferred to videos.
Authors: Lovis Heindrich, Falk Lieder
Abstract: The decisions of individuals and organizations are often suboptimal because fully rational decision-making is too demanding in the real world. Recent work suggests that some errors can be prevented by leveraging artificial intelligence to discover and teach clever heuristics. So far, this line of research has been limited to simplified, artificial decision-making tasks. This article is the first to extend this approach to a real-world decision problem, namely, executives deciding which project their organization should launch next. We develop a computational method (MGPS) that automatically discovers project selection strategies that are optimized for real people, and we develop an intelligent tutor that teaches the discovered project selection procedures. We evaluated MGPS on a computational benchmark and tested the intelligent tutor in a training experiment with two control conditions. MGPS outperformed a state-of-the-art method and was more computationally efficient. Moreover, people who practiced with our intelligent tutor learned significantly better project selection strategies than the control groups. These findings suggest that AI could be used to automate the process of discovering and formalizing the cognitive strategies taught by intelligent tutoring systems.
Authors: Yuanjian Xu, Anxian Liu, Jianing Hao, Zhenzhuo Li, Shichang Meng, Guang Zhang
Abstract: Modeling large-scale time series has gained significant attention in recent years. However, its direct application in finance remains challenging due to substantial differences in data characteristics across domains. Specifically, financial systems feature inherent stochasticity and low signal-to-noise ratios, rendering traditional methods and pre-training approaches ineffective. This underscores the urgent need for a foundation model tailored to financial time series. To bridge this gap, we propose \textbf{LENS}, a pre-trained model for this domain. \textbf{LENS} effectively captures the complexity of financial stochastic systems through a carefully crafted model architecture and mitigates noise during pre-training by using an invertible embedding module. We provide a rigorous theoretical explanation of the model's effectiveness and validate its performance through extensive experiments. Pre-trained on a dataset comprising 100 billion financial observations, \textbf{LENS} achieves exceptional results across a wide range of critical downstream tasks. Moreover, our work offers practical insights into developing pre-trained time series models in high-noise environments, paving the way for further advancements in this pivotal research domain.
Authors: Stelios Triantafyllou, Aleksa Sukovic, Yasaman Zolfimoselo, Goran Radanovic
Abstract: We address the challenge of explaining counterfactual outcomes in multi-agent Markov decision processes. In particular, we aim to explain the total counterfactual effect of an agent's action on the outcome of a realized scenario through its influence on the environment dynamics and the agents' behavior. To achieve this, we introduce a novel causal explanation formula that decomposes the counterfactual effect by attributing to each agent and state variable a score reflecting their respective contributions to the effect. First, we show that the total counterfactual effect of an agent's action can be decomposed into two components: one measuring the effect that propagates through all subsequent agents' actions and another related to the effect that propagates through the state transitions. Building on recent advancements in causal contribution analysis, we further decompose these two effects as follows. For the former, we consider agent-specific effects -- a causal concept that quantifies the counterfactual effect of an agent's action that propagates through a subset of agents. Based on this notion, we use Shapley value to attribute the effect to individual agents. For the latter, we consider the concept of structure-preserving interventions and attribute the effect to state variables based on their "intrinsic" contributions. Through extensive experimentation, we demonstrate the interpretability of our approach in a Gridworld environment with LLM-assisted agents and a sepsis management simulator.
Authors: Zijian Wu, Suozhi Huang, Zhejian Zhou, Huaiyuan Ying, Zheng Yuan, Wenwei Zhang, Dahua Lin, Kai Chen
Abstract: Large Language Models (LLMs) have emerged as powerful tools in mathematical theorem proving, particularly when utilizing formal languages such as LEAN. A prevalent proof method involves the LLM prover iteratively constructing the proof tactic by tactic, typically following a best-first search scheme. However, this method often ignores the critical preference information inside the existing tactic trajectories, hindering the search for deeper proofs. We propose an intuitive yet effective method, which utilizes a critic model to capture the preference information and to guide the search of the prover model at runtime. Given the prover-critic framework, a large-scale expert iteration with more than 20,000 CPU days is then applied to further fine-tune the prover and the critic. The trained InternLM2.5-StepProver critic significantly boosts the performance of the prover model (59.4% to 65.9%). We also analyze the impact of the critic on various aspects of the theorem proving process during expert iteration, providing insights into its effectiveness. We open-source our models and searched proofs at https://github.com/InternLM/InternLM-Math and https://huggingface.co/datasets/internlm/Lean-Workbook.
URLs: https://github.com/InternLM/InternLM-Math, https://huggingface.co/datasets/internlm/Lean-Workbook.
Authors: Mustafa O. Karabag, Jan Sobotka, Ufuk Topcu
Abstract: Large language model-based (LLM-based) agents have become common in settings that include non-cooperative parties. In such settings, agents' decision-making needs to conceal information from their adversaries, reveal information to their cooperators, and infer information to identify the other agents' characteristics. To investigate whether LLMs have these information control and decision-making capabilities, we make LLM agents play the language-based hidden-identity game, The Chameleon. In this game, a group of non-chameleon agents who do not know each other aim to identify the chameleon agent without revealing a secret. The game requires the aforementioned information control capabilities both as a chameleon and a non-chameleon. We begin with a theoretical analysis for a spectrum of strategies, from concealing to revealing, and provide bounds on the non-chameleons' winning probability. The empirical results with GPT, Gemini 2.5 Pro, Llama 3.1, and Qwen3 models show that while non-chameleon LLM agents identify the chameleon, they fail to conceal the secret from the chameleon, and their winning probability is far from the levels of even trivial strategies. Based on these empirical results and our theoretical analysis, we deduce that LLM-based agents may reveal excessive information to agents of unknown identities. Interestingly, we find that, when instructed to adopt an information-revealing level, this level is linearly encoded in the LLM's internal representations. While the instructions alone are often ineffective at making non-chameleon LLMs conceal, we show that steering the internal representations in this linear direction directly can reliably induce concealing behavior.
Authors: Leon Lang, Patrick Forr\'e
Abstract: As AI systems advance beyond human capabilities, scalable oversight becomes critical: how can we supervise AI that exceeds our abilities? A key challenge is that human evaluators may form incorrect beliefs about AI behavior in complex tasks, leading to unreliable feedback and poor value inference. To address this, we propose modeling evaluators' beliefs to interpret their feedback more reliably. We formalize human belief models, analyze their theoretical role in value learning, and characterize when ambiguity remains. To reduce reliance on precise belief models, we introduce "belief model covering" as a relaxation. This motivates our preliminary proposal to use the internal representations of adapted foundation models to mimic human evaluators' beliefs. These representations could be used to learn correct values from human feedback even when evaluators misunderstand the AI's behavior. Our work suggests that modeling human beliefs can improve value learning and outlines practical research directions for implementing this approach to scalable oversight.
Authors: Corina Catarau-Cotutiu, Esther Mondragon, Eduardo Alonso
Abstract: The ability of artificial intelligence agents to make optimal decisions and generalise them to different domains and tasks is compromised in complex scenarios. One way to address this issue has focused on learning efficient representations of the world and on how the actions of agents affect them in state-action transitions. Whereas such representations are procedurally efficient, they lack structural richness. To address this problem, we propose to enhance the agent's ontology and extend the traditional conceptualisation of trajectories to provide a more nuanced view of task execution. Structurally Enriched Trajectories (SETs) extend the encoding of sequences of states and their transitions by incorporating hierarchical relations between objects, interactions, and affordances. SETs are built as multi-level graphs, providing a detailed representation of the agent dynamics and a transferable functional abstraction of the task. SETs are integrated into an architecture, Structurally Enriched Trajectory Learning and Encoding (SETLE), that employs a heterogeneous graph-based memory structure of multi-level relational dependencies essential for generalisation. We demonstrate that SETLE can support downstream tasks, enabling agents to recognise task relevant structural patterns across CREATE and MiniGrid environments. Finally, we integrate SETLE with reinforcement learning and show measurable improvements in downstream performance, including breakthrough success rates in complex, sparse-reward tasks.
Authors: Haoran Luo, Haihong E, Guanting Chen, Yandan Zheng, Xiaobao Wu, Yikai Guo, Qika Lin, Yu Feng, Zemin Kuang, Meina Song, Yifan Zhu, Luu Anh Tuan
Abstract: Standard Retrieval-Augmented Generation (RAG) relies on chunk-based retrieval, whereas GraphRAG advances this approach by graph-based knowledge representation. However, existing graph-based RAG approaches are constrained by binary relations, as each edge in an ordinary graph connects only two entities, limiting their ability to represent the n-ary relations (n >= 2) in real-world knowledge. In this work, we propose HyperGraphRAG, a novel hypergraph-based RAG method that represents n-ary relational facts via hyperedges, and consists of knowledge hypergraph construction, retrieval, and generation. Experiments across medicine, agriculture, computer science, and law demonstrate that HyperGraphRAG outperforms both standard RAG and previous graph-based RAG methods in answer accuracy, retrieval efficiency, and generation quality. Our data and code are publicly available at https://github.com/LHRLAB/HyperGraphRAG.
Authors: Paresh Chaudhary, Yancheng Liang, Daphne Chen, Simon S. Du, Natasha Jaques
Abstract: Being able to cooperate with diverse humans is an important component of many economically valuable AI tasks, from household robotics to autonomous driving. However, generalizing to novel humans requires training on data that captures the diversity of human behaviors. Adversarial training is a promising method that allows dynamic data generation and ensures that agents are robust. It creates a feedback loop where the agent's performance influences the generation of new adversarial data, which can be used immediately to train the agent. However, adversarial training is difficult to apply in a cooperative task; how can we train an adversarial cooperator? We propose a novel strategy that combines a pretrained generative model to simulate valid cooperative agent policies with adversarial training to maximize regret. We call our method GOAT: Generative Online Adversarial Training. In this framework, the GOAT dynamically searches the latent space of the generative model for coordination strategies where the learning policy, the Cooperator agent, underperforms. GOAT enables better generalization by exposing the Cooperator to various challenging interaction scenarios. We maintain realistic coordination strategies by keeping the generative model frozen, thus avoiding adversarial exploitation. We evaluate GOAT with real human partners, and the results demonstrate state of the art performance on the Overcooked benchmark, highlighting its effectiveness in generalizing to diverse human behaviors.
Authors: Jongchan Park, Mingyu Park, Donghwan Lee
Abstract: Offline reinforcement learning (RL) aims to learn a policy from a static dataset without further interactions with the environment. Collecting sufficiently large datasets for offline RL is exhausting since this data collection requires colossus interactions with environments and becomes tricky when the interaction with the environment is restricted. Hence, how an agent learns the best policy with a minimal static dataset is a crucial issue in offline RL, similar to the sample efficiency problem in online RL. In this paper, we propose a simple yet effective plug-and-play pretraining method to initialize a feature of a Q-network to enhance data efficiency in offline RL. Specifically, we introduce a shared Q-network structure that outputs predictions of the next state and Q-value. We pretrain the shared Q-network through a supervised regression task that predicts a next state and trains the shared Q-network using diverse offline RL methods. Through extensive experiments, we empirically demonstrate that our method enhances the performance of existing popular offline RL methods on the D4RL, Robomimic and V-D4RL benchmarks. Furthermore, we show that our method significantly boosts data-efficient offline RL across various data qualities and data distributions trough D4RL and ExoRL benchmarks. Notably, our method adapted with only 10% of the dataset outperforms standard algorithms even with full datasets.
Authors: Geigh Zollicoffer, Minh Vu, Manish Bhattarai
Abstract: Vision-language models (VLMs) now rival human performance on many multimodal tasks, yet they still hallucinate objects or generate unsafe text. Current hallucination detectors, e.g., single-token linear probing (LP) and PTrue, typically analyze only the logit of the first generated token or just its highest-scoring component, overlooking richer signals embedded within earlier token distributions. We demonstrate that analyzing the complete sequence of early logits potentially provides substantially more diagnostic information. We emphasize that hallucinations may only emerge after several tokens, as subtle inconsistencies accumulate over time. By analyzing the Kullback-Leibler (KL) divergence between logits corresponding to hallucinated and non-hallucinated tokens, we underscore the importance of incorporating later-token logits to more accurately capture the reliability dynamics of VLMs. In response, we introduce Multi-Token Reliability Estimation (MTRE), a lightweight, white-box method that aggregates logits from the first ten tokens using multi-token log-likelihood ratios and self-attention. Despite the challenges posed by large vocabulary sizes and long logit sequences, MTRE remains efficient and tractable. Across MAD-Bench, MM-SafetyBench, MathVista, and four compositional-geometry benchmarks, MTRE achieves a 9.4% gain in accuracy and a 14.8% gain in AUROC over standard detection methods, establishing a new state of the art in hallucination detection for open-source VLMs.
Authors: Yuncheng Hua, Sion Weatherhead, Mehdi Jafari, Jianxiang Xie, Ji Miao, Hao Xue, Flora D. Salim
Abstract: Building credible simulators from data is difficult because structure design, parameter calibration, and out-of-distribution (OOD) robustness are tightly coupled. We introduce SOCIA (Simulation Orchestration for Computational Intelligence with Agents), a framework that treats simulator construction as joint structure-parameter co-optimization: it elicits mechanism-rich blueprints, exposes explicit tunable parameters, and instantiates a calibration schema, producing an executable simulator with built-in calibration hooks. SOCIA couples Bayesian Optimization for sample-efficient point calibration with Simulation-Based Inference for uncertainty-aware fitting; diagnostics trigger targeted structural edits in an outer refinement loop to co-optimize design and parameters under tight budgets. Across three diverse tasks, SOCIA consistently outperforms strong baselines, excelling on both in-distribution (ID) fitting and OOD shift. Ablations that weaken structure, calibration design, or tuning yield near-monotone degradations, underscoring the necessity of unified structure-parameter optimization. We will release the code soon.
Authors: Alfin Wijaya Rahardja, Junwei Liu, Weitong Chen, Zhenpeng Chen, Yiling Lou
Abstract: LLM-based agent systems are emerging as a new software paradigm and have been widely adopted across diverse domains such as medicine, robotics, and programming. However, maintaining these systems requires substantial effort, as they are inevitably prone to bugs and continually evolve to meet changing external requirements. Therefore, automatically resolving agent issues (i.e., bug reports or feature requests) is a crucial and challenging task. While recent software engineering (SE) agents (e.g., SWE-agent) have shown promise in addressing issues in traditional software systems, it remains unclear how effectively they can resolve real-world issues in agent systems, which differ significantly from traditional software. To fill this gap, we first manually analyze 201 real-world agent issues and identify common categories of agent issues. We then spend 500 person-hours constructing AGENTISSUE-BENCH, a reproducible benchmark comprising 50 agent issue resolution tasks (each with an executable environment and failure-triggering tests). We further evaluate state-of-the-art SE agents on AGENTISSUE-BENCH and reveal their limited effectiveness (i.e., with only 3.33% - 12.67% resolution rates). These results underscore the unique challenges of maintaining agent systems compared to traditional software, highlighting the need for further research to develop advanced SE agents for resolving agent issues. Data and code are available at https://alfin06.github.io/AgentIssue-Bench-Leaderboard/#/ .
URLs: https://alfin06.github.io/AgentIssue-Bench-Leaderboard/
Authors: Li Kang, Xiufeng Song, Heng Zhou, Yiran Qin, Jie Yang, Xiaohong Liu, Philip Torr, Lei Bai, Zhenfei Yin
Abstract: Coordinating multiple embodied agents in dynamic environments remains a core challenge in artificial intelligence, requiring both perception-driven reasoning and scalable cooperation strategies. While recent works have leveraged large language models (LLMs) for multi-agent planning, a few have begun to explore vision-language models (VLMs) for visual reasoning. However, these VLM-based approaches remain limited in their support for diverse embodiment types. In this work, we introduce VIKI-Bench, the first hierarchical benchmark tailored for embodied multi-agent cooperation, featuring three structured levels: agent activation, task planning, and trajectory perception. VIKI-Bench includes diverse robot embodiments, multi-view visual observations, and structured supervision signals to evaluate reasoning grounded in visual inputs. To demonstrate the utility of VIKI-Bench, we propose VIKI-R, a two-stage framework that fine-tunes a pretrained vision-language model (VLM) using Chain-of-Thought annotated demonstrations, followed by reinforcement learning under multi-level reward signals. Our extensive experiments show that VIKI-R significantly outperforms baselines method across all task levels. Furthermore, we show that reinforcement learning enables the emergence of compositional cooperation patterns among heterogeneous agents. Together, VIKI-Bench and VIKI-R offer a unified testbed and method for advancing multi-agent, visual-driven cooperation in embodied AI systems.
Authors: Khurram Yamin, Gaurav Ghosal, Bryan Wilder
Abstract: Large Language Models have been shown to contain extensive world knowledge in their parameters, enabling impressive performance on many knowledge intensive tasks. However, when deployed in novel settings, LLMs often encounter situations where they must integrate parametric knowledge with new or unfamiliar information. In this work, we explore whether LLMs can combine knowledge in-context with their parametric knowledge through the lens of counterfactual reasoning. Through synthetic and real experiments in multi-hop reasoning problems, we show that LLMs generally struggle with counterfactual reasoning, often resorting to exclusively using their parametric knowledge. Moreover, we show that simple post-hoc finetuning can struggle to instill counterfactual reasoning ability -- often leading to degradation in stored parametric knowledge. Ultimately, our work reveals important limitations of current LLM's abilities to re-purpose parametric knowledge in novel settings.
Authors: Zewen Liu, Juntong Ni, Xianfeng Tang, Max S. Y. Lau, Wenpeng Yin, Wei Jin
Abstract: Uncovering hidden symbolic laws from time series data, as an aspiration dating back to Kepler's discovery of planetary motion, remains a core challenge in scientific discovery and artificial intelligence. While Large Language Models show promise in structured reasoning tasks, their ability to infer interpretable, context-aligned symbolic structures from time series data is still underexplored. To systematically evaluate this capability, we introduce SymbolBench, a comprehensive benchmark designed to assess symbolic reasoning over real-world time series across three tasks: multivariate symbolic regression, Boolean network inference, and causal discovery. Unlike prior efforts limited to simple algebraic equations, SymbolBench spans a diverse set of symbolic forms with varying complexity. We further propose a unified framework that integrates LLMs with genetic programming to form a closed-loop symbolic reasoning system, where LLMs act both as predictors and evaluators. Our empirical results reveal key strengths and limitations of current models, highlighting the importance of combining domain knowledge, context alignment, and reasoning structure to improve LLMs in automated scientific discovery.
Authors: Meiping Wang, Jian Zhong, Rongduo Han, Liming Kang, Zhengkun Shi, Xiao Liang, Xing Lin, Nan Gao, Haining Zhang
Abstract: With the rapid development of mobile intelligent assistant technologies, multi-modal AI assistants have become essential interfaces for daily user interactions. However, current evaluation methods face challenges including high manual costs, inconsistent standards, and subjective bias. This paper proposes an automated multi-modal evaluation framework based on large language models and multi-agent collaboration. The framework employs a three-tier agent architecture consisting of interaction evaluation agents, semantic verification agents, and experience decision agents. Through supervised fine-tuning on the Qwen3-8B model, we achieve a significant evaluation matching accuracy with human experts. Experimental results on eight major intelligent agents demonstrate the framework's effectiveness in predicting users' satisfaction and identifying generation defects.
Authors: Hanyu Lai, Xiao Liu, Yanxiao Zhao, Han Xu, Hanchen Zhang, Bohao Jing, Yanyu Ren, Shuntian Yao, Yuxiao Dong, Jie Tang
Abstract: We introduce ComputerRL, a framework for autonomous desktop intelligence that enables agents to operate complex digital workspaces skillfully. ComputerRL features the API-GUI paradigm, which unifies programmatic API calls and direct GUI interaction to address the inherent mismatch between machine agents and human-centric desktop environments. Scaling end-to-end RL training is crucial for improvement and generalization across diverse desktop tasks; however, it remains challenging due to environmental inefficiency and instability during extended training. To support scalable and robust training, we develop a distributed RL infrastructure capable of orchestrating thousands of parallel virtual desktop environments to accelerate large-scale online RL. Furthermore, we propose Entropulse, a training strategy that alternates reinforcement learning with supervised fine-tuning, effectively mitigating entropy collapse during extended training runs. We employ ComputerRL on open models GLM-4-9B-0414 and GLM-4.1V-9B-Thinking, and evaluate them on the OSWorld benchmark. The AutoGLM-OS-9B achieves a new state-of-the-art accuracy of 48.9%, demonstrating significant improvements for general agents in desktop automation. Our code and the new OfficeWorld benchmark are available at https://github.com/thudm/ComputerRL. The algorithm and framework are adopted in building AutoGLM (Liu et al., 2024b).
Authors: Emmanuel O. Badmus, Peng Sang, Dimitrios Stamoulis, Amritanshu Pandey
Abstract: Rapid electrification and decarbonization are increasing the complexity of distribution grid (DG) operation and planning, necessitating advanced computational analyses to ensure reliability and resilience. These analyses depend on disparate workflows comprising complex models, function calls, and data pipelines that require substantial expert knowledge and remain difficult to automate. Workforce and budget constraints further limit utilities' ability to apply such analyses at scale. To address this gap, we build an agentic system PowerChain, which is capable of autonomously performing complex grid analyses. Existing agentic AI systems are typically developed in a bottom-up manner with customized context for predefined analysis tasks; therefore, they do not generalize to tasks that the agent has never seen. In comparison, to generalize to unseen DG analysis tasks, PowerChain dynamically generates structured context by leveraging supervisory signals from self-contained power systems tools (e.g., GridLAB-D) and an optimized set of expert-annotated and verified reasoning trajectories. For complex DG tasks defined in natural language, empirical results on real utility data demonstrate that PowerChain achieves up to a 144/% improvement in performance over baselines.
Authors: Shubham Gandhi, Jason Tsay, Jatin Ganhotra, Kiran Kate, Yara Rizk
Abstract: Large Language Model (LLM) agents are increasingly deployed for complex, multi-step software engineering (SWE) tasks. However, their trajectories often contain costly inefficiencies, such as redundant exploration, looping, and failure to terminate once a solution is reached. Prior work has largely treated these errors in a post-hoc manner, diagnosing failures only after execution. In this paper, we introduce SWE-PRM, an inference-time Process Reward Model (PRM) that intervenes during execution to detect and course-correct trajectory-level errors. Our PRM design leverages a taxonomy of common inefficiencies and delivers lightweight, interpretable feedback without modifying the underlying policy. On SWE-bench Verified, closed-source PRMs improve resolution from 40.0% to 50.6% (+10.6 p.p.), with the largest gains on medium and hard tasks. Among feedback strategies, taxonomy-guided PRMs outperform unguided or explicit action-prescriptive variants, increasing success rate while reducing trajectory length. These benefits come at an acceptable added inference cost of as low as $0.2, making PRMs a practical and scalable mechanism for improving SWE agents' reliability and efficiency.
Authors: Manvi Jha, Jiaxin Wan, Deming Chen
Abstract: Large Language Models (LLMs) have demonstrated impressive capabilities in automated code generation but frequently produce code that fails formal verification, an essential requirement for hardware and safety-critical domains. To overcome this fundamental limitation, we previously proposed PREFACE, a model-agnostic framework based on reinforcement learning (RL) that iteratively repairs the prompts provided to frozen LLMs, systematically steering them toward generating formally verifiable Dafny code without costly fine-tuning. This work presents Proof2Silicon, a novel end-to-end synthesis framework that embeds the previously proposed PREFACE flow to enable the generation of correctness-by-construction hardware directly from natural language specifications. Proof2Silicon operates by: (1) leveraging PREFACE's verifier-driven RL agent to optimize prompt generation iteratively, ensuring Dafny code correctness; (2) automatically translating verified Dafny programs into synthesizable high-level C using Dafny's Python backend and PyLog; and (3) employing Vivado HLS to produce RTL implementations. Evaluated rigorously on a challenging 100-task benchmark, PREFACE's RL-guided prompt optimization consistently improved Dafny verification success rates across diverse LLMs by up to 21%. Crucially, Proof2Silicon achieved an end-to-end hardware synthesis success rate of up to 72%, generating RTL designs through Vivado HLS synthesis flows. These results demonstrate a robust, scalable, and automated pipeline for LLM-driven, formally verified hardware synthesis, bridging natural-language specification and silicon realization.
Authors: Song Yu, Xiaofei Xu, Ke Deng, Li Li, Lin Tian
Abstract: Large language models (LLMs) face persistent challenges when handling long-context tasks, most notably the lost in the middle issue, where information located in the middle of a long input tends to be underutilized. Some existing methods that reduce input have the risk of discarding key information, while others that extend context windows often lead to attention dispersion. To address these limitations, we propose Tree of Agents (TOA), a multi-agent reasoning framework that segments the input into chunks processed by independent agents. Each agent generates its local cognition, then agents dynamically exchange information for collaborative reasoning along tree-structured paths. TOA enables agents to probe different reasoning orders for multi-perspective understanding, effectively mitigating position bias and reducing hallucinations. To improve processing efficiency, we incorporate prefix-hash caching and adaptive pruning strategies, achieving significant performance improvements with comparable API overhead. Experiments show that TOA, powered by compact LLaMA3.1-8B, significantly outperforms multiple baselines and demonstrates comparable performance to the latest and much larger commercial models, such as Gemini1.5-pro, on various long-context tasks. Code is available at https://github.com/Aireduce952/Tree-of-Agents.
Authors: Vincent Siu, Nathan W. Henry, Nicholas Crispino, Yang Liu, Dawn Song, Chenguang Wang
Abstract: While activation steering in large language models (LLMs) is a growing area of research, methods can often incur broader effects than desired. This motivates isolation of purer concept vectors to enable targeted interventions and understand LLM behavior at a more granular level. We present RepIt, a simple and data-efficient framework for isolating concept-specific representations. Across five frontier LLMs, RepIt enables precise interventions: it selectively suppresses refusal on targeted concepts while preserving refusal elsewhere, producing models that answer WMD-related questions while still scoring as safe on standard benchmarks. We further show that the corrective signal localizes to just 100-200 neurons and that robust target representations can be extracted from as few as a dozen examples on a single A6000. This efficiency raises a dual concern: manipulations can be performed with modest compute and data to extend to underrepresented data-scarce topics while evading existing benchmarks. By disentangling refusal vectors with RepIt, this work demonstrates that targeted interventions can counteract overgeneralization, laying the foundation for more granular control of model behavior.
Authors: Jiahao Yu, Zelei Cheng, Xian Wu, Xinyu Xing
Abstract: Large language models (LLMs) are increasingly used in various domains, showing impressive potential on different tasks. Recently, reasoning LLMs have been proposed to improve the \textit{reasoning} or \textit{thinking} capabilities of LLMs to solve complex problems. Despite the promising results of reasoning LLMs, enhancing the multi-step reasoning capabilities of LLMs still remains a significant challenge. While existing optimization methods have advanced the LLM reasoning capabilities, they often treat reasoning trajectories as a whole, without considering the underlying critical steps within the trajectory. In this paper, we introduce \textbf{G}uided \textbf{P}ivotal \textbf{O}ptimization (GPO), a novel fine-tuning strategy that dives into the reasoning process to enable more effective improvements. GPO first identifies the `critical step' within a reasoning trajectory - a point that the model must carefully proceed to succeed at the problem. We locate the critical step by estimating the advantage function. GPO then resets the policy to the critical step, samples the new rollout and prioritizes the learning process on those rollouts. This focus allows the model to learn more effectively from pivotal moments within the reasoning process to improve the reasoning performance. We demonstrate that GPO is a general strategy that can be integrated with various optimization methods to improve reasoning performance. Besides theoretical analysis, our experiments across challenging reasoning benchmarks show that GPO can consistently and significantly enhance the performance of existing optimization methods, showcasing its effectiveness and generalizability in improving LLM reasoning by concentrating on pivotal moments within the generation process.
Authors: Kang-il Lee, Jahyun Koo, Seunghyun Yoon, Minbeom Kim, Hyukhun Koh, Dongryeol Lee, Kyomin Jung
Abstract: We introduce transductive program synthesis, a new formulation of the program synthesis task that explicitly leverages test inputs during synthesis. While prior approaches to program synthesis--whether based on natural language descriptions or input-output examples--typically aim to generalize from training examples, they often struggle with robustness, especially in real-world settings where training examples are limited and test inputs involve various edge cases. To address this, we propose a novel framework that improves robustness by treating synthesis as an active learning over a finite hypothesis class defined by programs' outputs. We use an LLM to predict outputs for selected test inputs and eliminate inconsistent hypotheses, where the inputs are chosen via a greedy maximin algorithm to minimize the number of LLM queries required. We evaluate our approach on four benchmarks: Playgol, MBPP+, 1D-ARC, and programmatic world modeling on MiniGrid. We demonstrate that our method significantly improves program synthesis in both accuracy and efficiency. We release our code at https://github.com/klee972/SYNTRA.
Authors: Rubing Yang, Huajun Bai, Song Liu, Guanghua Yu, Runzhi Fan, Yanbin Dang, Jiejing Zhang, Kai Liu, Jianchen Zhu, Peng Chen
Abstract: Despite their strong performance on reasoning tasks, large reasoning models (LRMs) often suffer from overthinking, producing unnecessarily long outputs and incurring high end-to-end latency, a significant limitation to their real-world deployment. To address overthinking, early-exit mechanisms have been proposed to terminate reasoning before typical completion, showing that this approach can effectively shorten generation length with minimal impact on accuracy. However, their reliance on probing mechanisms introduces a detection overhead that limits their end-to-end latency gains and compromises their generalizability across diverse problems. Inspired by the use of hidden states in speculative decoding, we propose SpecExit, a novel framework that predicts both future tokens and an early-exit signal directly from a lightweight draft model without probing overhead. Our method offers significant improvements, reducing average generation length by 66\% and achieving a 2.5x speedup in end-to-end latency compared to the speculative decoding baseline, without compromising accuracy. Our method leverages the inherent signals from hidden states to provide effective early-exit signals, suggesting broader use of hidden states for efficient reasoning. Our code is available at https://github.com/Tencent/AngelSlim.
Authors: Yi Lu, Jianing Wang, Linsen Guo, Wei He, Hongyin Tang, Tao Gui, Xuanjing Huang, Xuezhi Cao, Wei Wang, Xunliang Cai
Abstract: Recent trends in test-time scaling for reasoning models (e.g., OpenAI o1, DeepSeek-R1) have led to remarkable improvements through long Chain-of-Thought (CoT). However, existing benchmarks mainly focus on immediate, single-horizon tasks, failing to adequately evaluate models' ability to understand and respond to complex, long-horizon scenarios. To address this incomplete evaluation of Large Reasoning Models (LRMs), we propose R-HORIZON, a method designed to stimulate long-horizon reasoning behaviors in LRMs through query composition. Based on R-HORIZON, we construct a long-horizon reasoning benchmark, comprising complex multi-step reasoning tasks with interdependent problems that span long reasoning horizons. Through comprehensive evaluation of LRMs using the R-HORIZON benchmark, we find that even the most advanced LRMs suffer significant performance degradation. Our analysis reveals that LRMs exhibit limited effective reasoning length and struggle to allocate thinking budget across multiple problems appropriately. Recognizing these limitations, we use R-HORIZON to construct long-horizon reasoning data for reinforcement learning with verified rewards (RLVR). Compared to training with single-horizon data, RLVR with R-HORIZON not only substantially improves performance on the multi-horizon reasoning tasks, but also promotes accuracy on standard reasoning tasks, with an increase of 7.5 on AIME2024. These results position R-HORIZON as a scalable, controllable, and low-cost paradigm for enhancing and evaluating the long-horizon reasoning capabilities of LRMs.
Authors: Marius Dragoi, Ioana Pintilie, Florin Gogianu, Florin Brad
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a powerful paradigm to improve Large Language Models on reasoning tasks such as coding, math or logic. To assess the reasoning boundary (the fraction of problems a model can solve) researchers often report Pass@k at large sampling budgets. Recent results reveal a crossover phenomenon: while RLVR models outperform the base model at small k values, the base model usually outperforms them when sampling a very large number of completions. This has been interpreted as evidence that base models have a larger reasoning boundary. We argue that on tasks with discrete answer spaces, such as math with numeric outputs, Pass@k at large k reflects the increasingly higher chance of success in the limit of the number of trials rather than genuine reasoning, and can therefore be misleading. We propose Cover@tau, which measures the fraction of problems that a model can solve for which at least a tau proportion of completions are correct. Unlike Pass@k, Cover@tau captures reasoning under an explicit reliability threshold: models that rely on random guessing degrade rapidly as tau increases. We evaluate several RLVR models using Cover@tau-based metrics and illustrate how the relative rankings of popular algorithms change compared to Pass@1, offering a different perspective on reasoning boundaries.
Authors: Qingni Wang, Yue Fan, Xin Eric Wang
Abstract: As large language models (LLMs) are increasingly deployed in risk-sensitive applications such as real-world open-ended question answering (QA), ensuring the trustworthiness of their outputs has become critical. Existing selective conformal prediction (SCP) methods provide statistical guarantees by constructing prediction sets with a constrained miscoverage rate for correct answers. However, prior works unrealistically assume that admissible answers for all instances can be obtained via finite sampling, even for open-ended QA scenarios that lack a fixed and finite solution space. To address this, we introduce a two-stage risk control framework comprising abstention-aware sampling and conformalized filtering (SAFER). Firstly, on a held-out calibration set, SAFER calibrates a sampling budget within the maximum sampling cap, using the Clopper-Pearson exact method at a user-desired risk level (i.e., the maximum allowable miscoverage rate of the sampling sets). If the risk level cannot be satisfied within the cap, we abstain; otherwise, the calibrated sampling budget becomes the minimum requirements at test time. Then, we employ calibration instances where correct answers are attainable under the calibrated budget and apply the conformal risk control method to determine a statistically valid uncertainty threshold, which filters unreliable distractors from the candidate set for each test data point. In this stage, SAFER introduces an additional risk level to guide the calculation of the threshold, thereby controlling the risk of correct answers being excluded. Furthermore, we show that SAFER is compatible with various task-specific admission criteria and calibration-test split ratios, highlighting its robustness and high data efficiency.
Authors: Trilok Padhi, Pinxian Lu, Abdulkadir Erol, Tanmay Sutar, Gauri Sharma, Mina Sonmez, Munmun De Choudhury, Ugur Kursuncu
Abstract: Large Language Model (LLM) agents are powering a growing share of interactive web applications, yet remain vulnerable to misuse and harm. Prior jailbreak research has largely focused on single-turn prompts, whereas real harassment often unfolds over multi-turn interactions. In this work, we present the Online Harassment Agentic Benchmark consisting of: (i) a synthetic multi-turn harassment conversation dataset, (ii) a multi-agent (e.g., harasser, victim) simulation informed by repeated game theory, (iii) three jailbreak methods attacking agents across memory, planning, and fine-tuning, and (iv) a mixed-methods evaluation framework. We utilize two prominent LLMs, LLaMA-3.1-8B-Instruct (open-source) and Gemini-2.0-flash (closed-source). Our results show that jailbreak tuning makes harassment nearly guaranteed with an attack success rate of 95.78--96.89% vs. 57.25--64.19% without tuning in Llama, and 99.33% vs. 98.46% without tuning in Gemini, while sharply reducing refusal rate to 1-2% in both models. The most prevalent toxic behaviors are Insult with 84.9--87.8% vs. 44.2--50.8% without tuning, and Flaming with 81.2--85.1% vs. 31.5--38.8% without tuning, indicating weaker guardrails compared to sensitive categories such as sexual or racial harassment. Qualitative evaluation further reveals that attacked agents reproduce human-like aggression profiles, such as Machiavellian/psychopathic patterns under planning, and narcissistic tendencies with memory. Counterintuitively, closed-source and open-source models exhibit distinct escalation trajectories across turns, with closed-source models showing significant vulnerability. Overall, our findings show that multi-turn and theory-grounded attacks not only succeed at high rates but also mimic human-like harassment dynamics, motivating the development of robust safety guardrails to ultimately keep online platforms safe and responsible.
Authors: Wangtao Sun, Xiang Cheng, Jialin Fan, Yao Xu, Xing Yu, Shizhu He, Jun Zhao, Kang Liu
Abstract: We study whether self-learning can scale LLM-based agents without relying on human-curated datasets or predefined rule-based rewards. Through controlled experiments in a search-agent setting, we identify two key determinants of scalable agent training: the source of reward signals and the scale of agent task data. We find that rewards from a Generative Reward Model (GRM) outperform rigid rule-based signals for open-domain learning, and that co-evolving the GRM with the policy further boosts performance. Increasing the volume of agent task data-even when synthetically generated-substantially enhances agentic capabilities. Building on these insights, we propose \textbf{Agentic Self-Learning} (ASL), a fully closed-loop, multi-role reinforcement learning framework that unifies task generation, policy execution, and evaluation within a shared tool environment and LLM backbone. ASL coordinates a Prompt Generator, a Policy Model, and a Generative Reward Model to form a virtuous cycle of harder task setting, sharper verification, and stronger solving. Empirically, ASL delivers steady, round-over-round gains, surpasses strong RLVR baselines (e.g., Search-R1) that plateau or degrade, and continues improving under zero-labeled-data conditions, indicating superior sample efficiency and robustness. We further show that GRM verification capacity is the main bottleneck: if frozen, it induces reward hacking and stalls progress; continual GRM training on the evolving data distribution mitigates this, and a small late-stage injection of real verification data raises the performance ceiling. This work establishes reward source and data scale as critical levers for open-domain agent learning and demonstrates the efficacy of multi-role co-evolution for scalable, self-improving agents. The data and code of this paper are released at https://github.com/forangel2014/Towards-Agentic-Self-Learning
URLs: https://github.com/forangel2014/Towards-Agentic-Self-Learning
Authors: Ruotian Peng, Yi Ren, Zhouliang Yu, Weiyang Liu, Yandong Wen
Abstract: Reinforcement learning with verifiable rewards (RLVR) has advanced the reasoning capabilities of large language models (LLMs). However, prevailing RLVR methods exhibit a systematic bias toward exploitation over exploration, as evidenced by improved pass@1 but reduced pass@K (K>1) performance. To understand this issue, we analyze training dynamics of RLVR methods by tracking the token-level probability distributions over vocabulary candidates. Our analysis reveals a consistent probability concentration effect where the top-1 candidate increasingly accumulates probability mass and suppresses that of other candidates. More importantly, stronger over-concentration correlates with worse pass@K performance. Inspired by this finding, we propose Simple Pass@K Optimization (SimKO), a method designed to mitigate the over-concentration issue, thereby encouraging exploration. SimKO operates in an asymmetrical manner. For verified-correct responses, it boosts the probabilities of the top-K candidates. For verified-incorrect responses, it applies stronger penalties to the top-1 candidate. We observe that this asymmetric design is particularly effective at mitigating over-concentration when applied at tokens with high entropy. Across various math and logical-reasoning benchmarks, SimKO consistently yields higher pass@K for a wide range of K, providing a simple way to improve RLVR's exploration.
Authors: Boyin Liu, Zhuo Zhang, Sen Huang, Lipeng Xie, Qingxu Fu, Haoran Chen, LI YU, Tianyi Hu, Zhaoyang Liu, Bolin Ding, Dongbin Zhao
Abstract: Aligning language models using LLM judge feedback offers a scalable alternative to human annotation, yet is plagued by judgment inconsistencies that destabilize reinforcement learning. While prior work has focused on judge accuracy, the critical issue of logical coherence particularly preference cycles has been largely unaddressed. To address this gap, this work introduces an end to end framework to systematically detect and resolve these inconsistencies within the reinforcement learning training loop. Our framework features two core contributions: the Conflict Detection Rate (CDR), a novel metric to quantify judgment conflicts, and Deconflicted Graph Rewards (DGR), a signal-purification framework that eliminates cycles before policy optimization. DGR constructs preference graphs from raw judgments, transforms them into conflict-free Directed Acyclic Graphs (DAGs), and generates a logically coherent reward signal compatible with any policy optimizer. Experiments confirm that our framework significantly improves training stability and model performance over strong baselines, establishing logical consistency as a crucial and now-addressable dimension of AI feedback. The code for our method is available at https://github.com/modelscope/RM-Gallery.
Authors: Yi Wan, Jiuqi Wang, Liam Li, Jinsong Liu, Ruihao Zhu, Zheqing Zhu
Abstract: Tool-augmented large language models (LLMs) are emerging as deep research agents, systems that decompose complex queries, retrieve external evidence, and synthesize grounded responses. Yet current agents remain limited by shallow retrieval, weak alignment metrics, and brittle tool-use behavior. We introduce PokeeResearch-7B, a 7B-parameter deep research agent built under a unified reinforcement learning framework for robustness, alignment, and scalability. PokeeResearch-7B is trained by an annotation-free Reinforcement Learning from AI Feedback (RLAIF) framework to optimize policies using LLM-based reward signals that capture factual accuracy, citation faithfulness, and instruction adherence. A chain-of-thought-driven multi-call reasoning scaffold further enhances robustness through self-verification and adaptive recovery from tool failures. Among 10 popular deep research benchmarks, PokeeResearch-7B achieves state-of-the-art performance among 7B-scale deep research agents. This highlights that careful reinforcement learning and reasoning design can produce efficient, resilient, and research-grade AI agents. The model and inference code is open-sourced under Apache 2.0 license at https://github.com/Pokee-AI/PokeeResearchOSS.
Authors: Tianxing Wu, Shutong Zhu, Jingting Wang, Ning Xu, Guilin Qi, Haofen Wang
Abstract: Uncertain knowledge graphs (UKGs) associate each triple with a confidence score to provide more precise knowledge representations. Recently, since real-world UKGs suffer from the incompleteness, uncertain knowledge graph (UKG) completion attracts more attention, aiming to complete missing triples and confidences. Current studies attempt to learn UKG embeddings to solve this problem, but they neglect the extremely imbalanced distributions of triple confidences. This causes that the learnt embeddings are insufficient to high-quality UKG completion. Thus, in this paper, to address the above issue, we propose a new semi-supervised Confidence Distribution Learning (ssCDL) method for UKG completion, where each triple confidence is transformed into a confidence distribution to introduce more supervision information of different confidences to reinforce the embedding learning process. ssCDL iteratively learns UKG embedding by relational learning on labeled data (i.e., existing triples with confidences) and unlabeled data with pseudo labels (i.e., unseen triples with the generated confidences), which are predicted by meta-learning to augment the training data and rebalance the distribution of triple confidences. Experiments on two UKG datasets demonstrate that ssCDL consistently outperforms state-of-the-art baselines in different evaluation metrics.
Authors: Melik Ozolcer, Sang Won Bae
Abstract: We study a web-deployed, tool-augmented LLM health coach with real users. In a pilot with seven users (280 rated turns), offline policy evaluation (OPE) over factorized decision heads (Tool/Style) shows that a uniform heavy-tool policy raises average value on logs but harms specific subgroups, most notably low-health-literacy/high-self-efficacy users. A lightweight simulator with hidden archetypes further shows that adding a small early information-gain bonus reliably shortens trait identification and improves goal success and pass@3. Together, these early findings indicate an evaluation-first path to personalization: freeze the generator, learn subgroup-aware decision heads on typed rewards (objective tool outcomes and satisfaction), and always report per-archetype metrics to surface subgroup harms that averages obscure.
Authors: Lyuzhou Chen, Taiyu Ban, Xiangyu Wang, Derui Lyu, Huanhuan Chen
Abstract: Causal structure learning (CSL), a prominent technique for encoding cause-and-effect relationships among variables, through Bayesian Networks (BNs). Although recovering causal structure solely from data is a challenge, the integration of prior knowledge, revealing partial structural truth, can markedly enhance learning quality. However, current methods based on prior knowledge exhibit limited resilience to errors in the prior, with hard constraint methods disregarding priors entirely, and soft constraints accepting priors based on a predetermined confidence level, which may require expert intervention. To address this issue, we propose a strategy resilient to edge-level prior errors for CSL, thereby minimizing human intervention. We classify prior errors into different types and provide their theoretical impact on the Structural Hamming Distance (SHD) under the presumption of sufficient data. Intriguingly, we discover and prove that the strong hazard of prior errors is associated with a unique acyclic closed structure, defined as ``quasi-circle''. Leveraging this insight, a post-hoc strategy is employed to identify the prior errors by its impact on the increment of ``quasi-circles''. Through empirical evaluation on both real and synthetic datasets, we demonstrate our strategy's robustness against prior errors. Specifically, we highlight its substantial ability to resist order-reversed errors while maintaining the majority of correct prior.
Authors: Chrisantus Eze, Christopher Crick
Abstract: Robot learning of manipulation skills is hindered by the scarcity of diverse, unbiased datasets. While curated datasets can help, challenges remain in generalizability and real-world transfer. Meanwhile, large-scale "in-the-wild" video datasets have driven progress in computer vision through self-supervised techniques. Translating this to robotics, recent works have explored learning manipulation skills by passively watching abundant videos sourced online. Showing promising results, such video-based learning paradigms provide scalable supervision while reducing dataset bias. This survey reviews foundations such as video feature representation learning techniques, object affordance understanding, 3D hand/body modeling, and large-scale robot resources, as well as emerging techniques for acquiring robot manipulation skills from uncontrolled video demonstrations. We discuss how learning only from observing large-scale human videos can enhance generalization and sample efficiency for robotic manipulation. The survey summarizes video-based learning approaches, analyses their benefits over standard datasets, survey metrics, and benchmarks, and discusses open challenges and future directions in this nascent domain at the intersection of computer vision, natural language processing, and robot learning.
Authors: Xue Jiang, Yihong Dong, Zhiyuan Fan, Zhi Jin, Wenpin Jiao, Ge Li
Abstract: Although Large Language Models (LLMs) have made significant progress in code generation, they still struggle with code generation tasks in specific scenarios. These scenarios usually necessitate the adaptation of LLMs to fulfill specific needs, but the limited training data available in practice leads to poor code generation performance. Therefore, how to effectively adapt LLMs to new scenarios with few training data is a major challenge for current code generation. In this paper, we propose a novel adaptation approach named DEED, which stands for Data-Efficient adaptation with Error-Driven learning for code generation. DEED leverages the errors made by LLMs as learning opportunities, using error revision to overcome their own shortcomings, thus achieving efficient learning. Specifically, DEED involves identifying error code generated by LLMs, employing Self-Revise for code revision, optimizing the model with revised code, and iteratively adapting the process for continuous improvement. Experimental results show that, compared to other mainstream fine-tuning approaches, DEED achieves superior performance with few training data, showing an average relative improvement of 46.2% in Pass@1 on multiple code generation benchmarks. We also validate the effectiveness of Self-Revise, which generates revised code that optimizes the model more efficiently compared to the code samples from datasets. Moreover, DEED consistently demonstrates strong performance across various LLMs, underscoring its applicability.
Authors: Siya Qi, Lin Gui, Yulan He, Zheng Yuan
Abstract: The rapid advancement of Large Language Models (LLMs) has brought a pressing challenge: how to reliably assess hallucinations to guarantee model trustworthiness. Although Automatic Hallucination Evaluation (AHE) has become an indispensable component of this effort, the field remains fragmented in its methodologies, limiting both conceptual clarity and practical progress. This survey addresses this critical gap through a systematic analysis of 105 evaluation methods, revealing that 77.1% specifically target LLMs, a paradigm shift that demands new evaluation frameworks. We formulate a structured framework to organize the field, based on a survey of foundational datasets and benchmarks and a taxonomy of evaluation methodologies, which together systematically document the evolution from pre-LLM to post-LLM approaches. Beyond taxonomical organization, we identify fundamental limitations in current approaches and their implications for real-world deployment. To guide future research, we delineate key challenges and propose strategic directions, including enhanced interpretability mechanisms and integration of application-specific evaluation criteria, ultimately providing a roadmap for developing more robust and practical hallucination evaluation systems.
Authors: Charmaine Barker, Daniel Bethell, Dimitar Kazakov
Abstract: Mitigating bias in automated decision-making systems, particularly in deep learning models, is a critical challenge due to nuanced definitions of fairness, dataset-specific biases, and the inherent trade-off between fairness and accuracy. To address these issues, we introduce FairVIC, an innovative approach that enhances fairness in neural networks by integrating variance, invariance, and covariance terms into the loss function during training. Unlike methods that rely on predefined fairness criteria, FairVIC abstracts fairness concepts to minimise dependency on protected characteristics. We evaluate FairVIC against comparable bias mitigation techniques on benchmark datasets, considering both group and individual fairness, and conduct an ablation study on the accuracy-fairness trade-off. FairVIC demonstrates significant improvements ($\approx70\%$) in fairness across all tested metrics without compromising accuracy, thus offering a robust, generalisable solution for fair deep learning across diverse tasks and datasets.
Authors: Thanet Markchom, Huizhi Liang, James Ferryman
Abstract: Explainability of recommender systems has become essential to ensure users' trust and satisfaction. Various types of explainable recommender systems have been proposed including explainable graph-based recommender systems. This review paper discusses state-of-the-art approaches of these systems and categorizes them based on three aspects: learning methods, explaining methods, and explanation types. It also explores the commonly used datasets, explainability evaluation methods, and future directions of this research area. Compared with the existing review papers, this paper focuses on explainability based on graphs and covers the topics required for developing novel explainable graph-based recommender systems.
Authors: Jiahao Yu, Xian Wu, Hao Liu, Wenbo Guo, Xinyu Xing
Abstract: We propose BlockScan, a customized Transformer for anomaly detection in blockchain transactions. Unlike existing methods that rely on rule-based systems or directly apply off-the-shelf large language models (LLMs), BlockScan introduces a series of customized designs to effectively model the unique data structure of blockchain transactions. First, a blockchain transaction is multi-modal, containing blockchain-specific tokens, texts, and numbers. We design a novel modularized tokenizer to handle these multi-modal inputs, balancing the information across different modalities. Second, we design a customized masked language modeling mechanism for pretraining the Transformer architecture, incorporating RoPE embedding and FlashAttention for handling longer sequences. Finally, we design a novel anomaly detection method based on the model outputs. We further provide theoretical analysis for the detection method of our system. Extensive evaluations on Ethereum and Solana transactions demonstrate BlockScan's exceptional capability in anomaly detection while maintaining a low false positive rate. Remarkably, BlockScan is the only method that successfully detects anomalous transactions on Solana with high accuracy, whereas all other approaches achieved very low or zero detection recall scores. This work sets a new benchmark for applying Transformer-based approaches in blockchain data analysis.
Authors: Attila Lovas
Abstract: Nonlinear time series models with exogenous regressors are essential in econometrics, queuing theory, and machine learning, though their statistical analysis remains incomplete. Key results, such as the law of large numbers and the functional central limit theorem, are known for weakly dependent variables. We demonstrate the transfer of mixing properties from the exogenous regressor to the response via coupling arguments. Additionally, we study Markov chains in random environments with drift and minorization conditions, even under non-stationary environments with favorable mixing properties, and apply this framework to single-server queuing models.
Authors: Zhijie Nie, Zhangchi Feng, Mingxin Li, Cunwang Zhang, Yanzhao Zhang, Dingkun Long, Richong Zhang
Abstract: Text embedding has become a foundational technology in natural language processing (NLP) during the deep learning era, driving advancements across a wide array of downstream tasks. While many natural language understanding challenges can now be modeled using generative paradigms and leverage the robust generative and comprehension capabilities of large language models (LLMs), numerous practical applications - such as semantic matching, clustering, and information retrieval - continue to rely on text embeddings for their efficiency and effectiveness. Therefore, integrating LLMs with text embeddings has become a major research focus in recent years. In this survey, we categorize the interplay between LLMs and text embeddings into three overarching themes: (1) LLM-augmented text embedding, enhancing traditional embedding methods with LLMs; (2) LLMs as text embedders, adapting their innate capabilities for high-quality embedding; and (3) Text embedding understanding with LLMs, leveraging LLMs to analyze and interpret embeddings. By organizing recent works based on interaction patterns rather than specific downstream applications, we offer a novel and systematic overview of contributions from various research and application domains in the era of LLMs. Furthermore, we highlight the unresolved challenges that persisted in the pre-LLM era with pre-trained language models (PLMs) and explore the emerging obstacles brought forth by LLMs. Building on this analysis, we outline prospective directions for the evolution of text embedding, addressing both theoretical and practical opportunities in the rapidly advancing landscape of NLP.
Authors: Chengrui Gao, Ziyuan Yang, Wei Jia, Lu Leng, Bob Zhang, Andrew Beng Jin Teoh
Abstract: Palmprint recognition has emerged as a prominent biometric technology, widely applied in diverse scenarios. Traditional handcrafted methods for palmprint recognition often fall short in representation capability, as they heavily depend on researchers' prior knowledge. Deep learning (DL) has been introduced to address this limitation, leveraging its remarkable successes across various domains. While existing surveys focus narrowly on specific tasks within palmprint recognition-often grounded in traditional methodologies-there remains a significant gap in comprehensive research exploring DL-based approaches across all facets of palmprint recognition. This paper bridges that gap by thoroughly reviewing recent advancements in DL-powered palmprint recognition. The paper systematically examines progress across key tasks, including region-of-interest segmentation, feature extraction, and security/privacy-oriented challenges. Beyond highlighting these advancements, the paper identifies current challenges and uncovers promising opportunities for future research. By consolidating state-of-the-art progress, this review serves as a valuable resource for researchers, enabling them to stay abreast of cutting-edge technologies and drive innovation in palmprint recognition.
Authors: Rajdeep Haldar, Ziyi Wang, Qifan Song, Guang Lin, Yue Xing
Abstract: We present a theoretical framework showing that popular LLM alignment methods, including RLHF and its variants, can be understood as divergence estimators between aligned (safe or preferred) and unaligned (harmful or less preferred) distributions. This perspective explains the emergence of separation in the latent space between safe and harmful prompts after alignment. As an application of our general divergence framework, we propose KLDO, a novel KL divergence-based alignment method, and empirically validate its effectiveness. We further show that using compliance-refusal datasets, rather than standard preference-based datasets, leads to stronger separation and improved safety alignment. Finally, to quantify the separation effect, we propose a distance-based metric in the prompt representation space, which also acts as a statistically significant indicator for model safety.
Authors: Jinwei Hu, Zhenglin Huang, Xiangyu Yin, Wenjie Ruan, Guangliang Cheng, Yi Dong, Xiaowei Huang
Abstract: Large language models have been widely applied, but can inadvertently encode sensitive or harmful information, raising significant safety concerns. Machine unlearning has emerged to alleviate this concern; however, existing training-time unlearning approaches, relying on coarse-grained loss combinations, have limitations in precisely separating knowledge and balancing removal effectiveness with model utility. In contrast, we propose Fine-grained Activation manipuLation by Contrastive Orthogonal uNalignment (FALCON), a novel representation-guided unlearning approach that leverages information-theoretic guidance for efficient parameter selection, employs contrastive mechanisms to enhance representation separation, and projects conflict gradients onto orthogonal subspaces to resolve conflicts between forgetting and retention objectives. Extensive experiments demonstrate that FALCON achieves superior unlearning effectiveness while maintaining model utility, exhibiting robust resistance against knowledge recovery attempts.
Authors: Zeki Doruk Erden, Boi Faltings
Abstract: Inherent limitations of contemporary machine learning systems in crucial areas -- importantly in continual learning, information reuse, comprehensibility, and integration with deliberate behavior -- are receiving increasing attention. To address these challenges, we introduce a system design, fueled by a novel learning approach conceptually grounded in principles of evolutionary developmental biology, that overcomes key limitations of current methods. Our design comprises three core components: The Modeller, a gradient-free learning mechanism inherently capable of continual learning and structural adaptation; a planner for goal-directed action over learned models; and a behavior encapsulation mechanism that can decompose complex behaviors into a hierarchical structure. We demonstrate proof-of-principle operation in a simple test environment. Additionally, we extend our modeling framework to higher-dimensional network-structured spaces, using MNIST for a shape detection task. Our framework shows promise in overcoming multiple major limitations of contemporary machine learning systems simultaneously and in an organic manner.
Authors: Felix Dobslaw, Robert Feldt, Juyeon Yoon, Shin Yoo
Abstract: Large Language Models (LLMs) and Multi-Agent LLMs (MALLMs) introduce non-determinism unlike traditional or machine learning software, requiring new approaches to verifying correctness beyond simple output comparisons or statistical accuracy over test datasets. This paper presents a taxonomy for LLM test case design, informed by research literature and our experience. Each facet is exemplified, and we conduct an LLM-assisted analysis of six open-source testing frameworks, perform a sensitivity study of an agent-based system across different model configurations, and provide working examples contrasting atomic and aggregated test cases. We identify key variation points that impact test correctness and highlight open challenges that the research, industry, and open-source communities must address as LLMs become integral to software systems. Our taxonomy defines four facets of LLM test case design, addressing ambiguity in both inputs and outputs while establishing best practices. It distinguishes variability in goals, the system under test, and inputs, and introduces two key oracle types: atomic and aggregated. Our findings reveal that current tools treat test executions as isolated events, lack explicit aggregation mechanisms, and inadequately capture variability across model versions, configurations, and repeated runs. This highlights the need for viewing correctness as a distribution of outcomes rather than a binary property, requiring closer collaboration between academia and practitioners to establish mature, variability-aware testing methodologies.
Authors: Xue Han, Qian Hu, Yitong Wang, Wenchun Gao, Lianlian Zhang, Qing Wang, Lijun Mei, Chao Deng, Junlan Feng
Abstract: Large language models (LLMs) suffer from temporal misalignment issues especially across long span of time. The issue arises from knowing that LLMs are trained on large amounts of data where temporal information is rather sparse over long times, such as thousands of years, resulting in insufficient learning or catastrophic forgetting by the LLMs. This paper proposes a methodology named "Ticktack" for addressing the LLM's long-time span misalignment in a yearly setting. Specifically, we first propose to utilize the sexagenary year expression instead of the Gregorian year expression employed by LLMs, achieving a more uniform distribution in yearly granularity. Then, we employ polar coordinates to model the sexagenary cycle of 60 terms and the year order within each term, with additional temporal encoding to ensure LLMs understand them. Finally, we present a temporal representational alignment approach for post-training LLMs that effectively distinguishes time points with relevant knowledge, hence improving performance on time-related tasks, particularly over a long period. We also create a long time span benchmark for evaluation. Experimental results prove the effectiveness of our proposal.
Authors: Nir Ailon, Akhiad Bercovich, Yahel Uffenheimer, Omri Weinstein
Abstract: Modern AI relies on huge matrix multiplications (MatMuls), whose computation poses a scalability problem for inference and training. We propose an alternative, GPU native bilinear operator to MatMuls in neural networks, which offers a three-way tradeoff between: speed, accuracy and parameter count. In particular, this operator requires substantially fewer FLOPs to evaluate ($\ll n^3$), yet increases the parameter count compared to MatMul ($\gg n^2$). We call this operator Strassen-Tile (STL). The key idea behind STL is a local learnable change-of-basis, applied on tiles of the weight and activation matrices, followed by an element-wise product between the tiles, implemented simultaneously via MatMul. The key technical question we study is how to optimize the change-of-basis of a given layer, which is a highly non-convex problem. We show that theory-backed initializations (inspired by fast matrix and polynomial multiplication) lead to substantially better accuracy than random SGD initialization. This phenomenon motivates further algorithmic study of STL optimization in DNNs. Our experiments demonstrate that STL can approximate 4x4 MatMul of tiles while reducing FLOPs by a factor of 2.66, and can improve Imagenet-1K accuracy of SoTA T2T-ViT-7 (4.3M parameters) while lowering FLOPs. Even with non-CUDA optimized PyTorch code, STL achieves wall-clock speedups in the compute-bound regime. These results, together with its theoretical grounds, suggest STL as a promising building block for scalable and cost-efficient AI.
Authors: Michal Balcerak, Tamaz Amiranashvili, Antonio Terpin, Suprosanna Shit, Lea Bogensperger, Sebastian Kaltenbach, Petros Koumoutsakos, Bjoern Menze
Abstract: Current state-of-the-art generative models map noise to data distributions by matching flows or scores. A key limitation of these models is their inability to readily integrate available partial observations and additional priors. In contrast, energy-based models (EBMs) address this by incorporating corresponding scalar energy terms. Here, we propose Energy Matching, a framework that endows flow-based approaches with the flexibility of EBMs. Far from the data manifold, samples move from noise to data along irrotational, optimal transport paths. As they approach the data manifold, an entropic energy term guides the system into a Boltzmann equilibrium distribution, explicitly capturing the underlying likelihood structure of the data. We parameterize these dynamics with a single time-independent scalar field, which serves as both a powerful generator and a flexible prior for effective regularization of inverse problems. The present method substantially outperforms existing EBMs on CIFAR-10 and ImageNet generation in terms of fidelity, while retaining simulation-free training of transport-based approaches away from the data manifold. Furthermore, we leverage the flexibility of the method to introduce an interaction energy that supports the exploration of diverse modes, which we demonstrate in a controlled protein generation setting. This approach learns a scalar potential energy, without time conditioning, auxiliary generators, or additional networks, marking a significant departure from recent EBM methods. We believe this simplified yet rigorous formulation significantly advances EBMs capabilities and paves the way for their wider adoption in generative modeling in diverse domains.
Authors: Long Li, Jiajia Li, Dong Chen, Lina Pu, Haibo Yao, Yanbo Huang
Abstract: In modern smart agriculture, object detection plays a crucial role by enabling automation, precision farming, and monitoring of resources. From identifying crop health and pest infestations to optimizing harvesting processes, accurate object detection enhances both productivity and sustainability. However, training object detection models often requires large-scale data collection and raises privacy concerns, particularly when sensitive agricultural data is distributed across farms. To address these challenges, we propose VLLFL, a vision-language model-based lightweight federated learning framework (VLLFL). It harnesses the generalization and context-aware detection capabilities of the vision-language model (VLM) and leverages the privacy-preserving nature of federated learning. By training a compact prompt generator to boost the performance of the VLM deployed across different farms, VLLFL preserves privacy while reducing communication overhead. Experimental results demonstrate that VLLFL achieves 14.53% improvement in the performance of VLM while reducing 99.3% communication overhead. Spanning tasks from identifying a wide variety of fruits to detecting harmful animals in agriculture, the proposed framework offers an efficient, scalable, and privacy-preserving solution specifically tailored to agricultural applications.
Authors: Yui Lo, Yuqian Chen, Dongnan Liu, Leo Zekelman, Jarrett Rushmore, Yogesh Rathi, Nikos Makris, Alexandra J. Golby, Fan Zhang, Weidong Cai, Lauren J. O'Donnell
Abstract: Shape measures have emerged as promising descriptors of white matter tractography, offering complementary insights into anatomical variability and associations with cognitive and clinical phenotypes. However, conventional methods for computing shape measures are computationally expensive and time-consuming for large-scale datasets due to reliance on voxel-based representations. We propose Tract2Shape, a novel multimodal deep learning framework that leverages geometric (point cloud) and scalar (tabular) features to predict ten white matter tractography shape measures. To enhance model efficiency, we utilize a dimensionality reduction algorithm for the model to predict five primary shape components. The model is trained and evaluated on two independently acquired datasets, the HCP-YA dataset, and the PPMI dataset. We evaluate the performance of Tract2Shape by training and testing it on the HCP-YA dataset and comparing the results with state-of-the-art models. To further assess its robustness and generalization ability, we also test Tract2Shape on the unseen PPMI dataset. Tract2Shape outperforms SOTA deep learning models across all ten shape measures, achieving the highest average Pearson's r and the lowest nMSE on the HCP-YA dataset. The ablation study shows that both multimodal input and PCA contribute to performance gains. On the unseen testing PPMI dataset, Tract2Shape maintains a high Pearson's r and low nMSE, demonstrating strong generalizability in cross-dataset evaluation. Tract2Shape enables fast, accurate, and generalizable prediction of white matter shape measures from tractography data, supporting scalable analysis across datasets. This framework lays a promising foundation for future large-scale white matter shape analysis.
Authors: A N M Nafiul Islam, Xuezhong Niu, Jiahui Duan, Shubham Kumar, Kai Ni, Abhronil Sengupta
Abstract: Although inspired by neuronal systems in the brain, artificial neural networks generally employ point-neurons, which offer far less computational complexity than their biological counterparts. Neurons have dendritic arbors that connect to different sets of synapses and offer local non-linear accumulation - playing a pivotal role in processing and learning. Inspired by this, we propose a novel neuron design based on a multi-gate ferroelectric field-effect transistor that mimics dendrites. It leverages ferroelectric nonlinearity for local computations within dendritic branches, while utilizing the transistor action to generate the final neuronal output. The branched architecture paves the way for utilizing smaller crossbar arrays in hardware integration, leading to greater efficiency. Using an experimentally calibrated device-circuit-algorithm co-simulation framework, we demonstrate that networks incorporating our dendritic neurons achieve superior performance in comparison to much larger networks without dendrites ($\sim$17$\times$ fewer trainable weight parameters). These findings suggest that dendritic hardware can significantly improve computational efficiency, and learning capacity of neuromorphic systems optimized for edge applications.
Authors: Sebastian Rassmann, David K\"ugler, Christian Ewert, Martin Reuter
Abstract: While Generative Adversarial Nets (GANs) and Diffusion Models (DMs) have achieved impressive results in natural image synthesis, their core strengths - creativity and realism - can be detrimental in medical applications, where accuracy and fidelity are paramount. These models instead risk introducing hallucinations and replication of unwanted acquisition noise. Here, we propose YODA (You Only Denoise once - or Average), a 2.5D diffusion-based framework for medical image translation (MIT). Consistent with DM theory, we find that conventional diffusion sampling stochastically replicates noise. To mitigate this, we draw and average multiple samples, akin to physical signal averaging. As this effectively approximates the DM's expected value, we term this Expectation-Approximation (ExpA) sampling. We additionally propose regression sampling YODA, which retains the initial DM prediction and omits iterative refinement to produce noise-free images in a single step. Across five diverse multi-modal datasets - including multi-contrast brain MRI and pelvic MRI-CT - we demonstrate that regression sampling is not only substantially more efficient but also matches or exceeds image quality of full diffusion sampling even with ExpA. Our results reveal that iterative refinement solely enhances perceptual realism without benefiting information translation, which we confirm in relevant downstream tasks. YODA outperforms eight state-of-the-art DMs and GANs and challenges the presumed superiority of DMs and GANs over computationally cheap regression models for high-quality MIT. Furthermore, we show that YODA-translated images are interchangeable with, or even superior to, physical acquisitions for several medical applications.
Authors: Zuwei Long, Yunhang Shen, Chaoyou Fu, Heting Gao, Lijiang Li, Peixian Chen, Mengdan Zhang, Hang Shao, Jian Li, Jinlong Peng, Haoyu Cao, Ke Li, Rongrong Ji, Xing Sun
Abstract: With the growing requirement for natural human-computer interaction, speech-based systems receive increasing attention as speech is one of the most common forms of daily communication. However, the existing speech models still experience high latency when generating the first audio token during streaming, which poses a significant bottleneck for deployment. To address this issue, we propose VITA-Audio, an end-to-end large speech model with fast audio-text token generation. Specifically, we introduce a lightweight Multiple Cross-modal Token Prediction (MCTP) module that efficiently generates multiple audio tokens within a single model forward pass, which not only accelerates the inference but also significantly reduces the latency for generating the first audio in streaming scenarios. In addition, a four-stage progressive training strategy is explored to achieve model acceleration with minimal loss of speech quality. To our knowledge, VITA-Audio is the first multi-modal large language model capable of generating audio output during the first forward pass, enabling real-time conversational capabilities with minimal latency. VITA-Audio is fully reproducible and is trained on open-source data only. Experimental results demonstrate that our model achieves an inference speedup of 3~5x at the 7B parameter scale, but also significantly outperforms open-source models of similar model size on multiple benchmarks for automatic speech recognition (ASR), text-to-speech (TTS), and spoken question answering (SQA) tasks.
Authors: Simon Suh
Abstract: The adoption of open science has quickly changed how artificial intelligence (AI) policy research is distributed globally. This study examines the regional trends in the citation of preprints, specifically focusing on the impact of two major disruptive events: the COVID-19 pandemic and the release of ChatGPT, on research dissemination patterns in the United States, Europe, and South Korea from 2015 to 2024. Using bibliometrics data from the Web of Science, this study tracks how global disruptive events influenced the adoption of preprints in AI policy research and how such shifts vary by region. By marking the timing of these disruptive events, the analysis reveals that while all regions experienced growth in preprint citations, the magnitude and trajectory of change varied significantly. The United States exhibited sharp, event-driven increases; Europe demonstrated institutional growth; and South Korea maintained consistent, linear growth in preprint adoption. These findings suggest that global disruptions may have accelerated preprint adoption, but the extent and trajectory are shaped by local research cultures, policy environments, and levels of open science maturity. This paper emphasizes the need for future AI governance strategies to consider regional variability in research dissemination and highlights opportunities for further longitudinal and comparative research to deepen our understanding of open-access adoption in AI policy development.
Authors: Yingli Shen, Wen Lai, Shuo Wang, Ge Gao, Kangyang Luo, Alexander Fraser, Maosong Sun
Abstract: Continued pretraining and instruction tuning on large-scale multilingual data have proven to be effective in scaling large language models (LLMs) to low-resource languages. However, the unaligned nature of such data limits its ability to effectively capture cross-lingual semantics. In contrast, multi-way parallel data, where identical content is aligned across multiple languages, provides stronger cross-lingual consistency and offers greater potential for improving multilingual performance. In this paper, we introduce a large-scale, high-quality multi-way parallel corpus, TED2025, based on TED Talks. The corpus spans 113 languages, with up to 50 languages aligned in parallel, ensuring extensive multilingual coverage. Using this dataset, we investigate best practices for leveraging multi-way parallel data to enhance LLMs, including strategies for continued pretraining, instruction tuning, and the analysis of key influencing factors. Experiments on six multilingual benchmarks show that models trained on multiway parallel data consistently outperform those trained on unaligned multilingual data.
Authors: Quan Nguyen, Thanh Nguyen-Tang
Abstract: We study the approximation capabilities, convergence speeds and on-convergence behaviors of transformers trained on in-context recall tasks -- which requires to recognize the \emph{positional} association between a pair of tokens from in-context examples. Existing theoretical results only focus on the in-context reasoning behavior of transformers after being trained for the \emph{one} gradient descent step. It remains unclear what is the on-convergence behavior of transformers being trained by gradient descent and how fast the convergence rate is. In addition, the generalization of transformers in one-step in-context reasoning has not been formally investigated. This work addresses these gaps. We first show that a class of transformers with either linear, ReLU or softmax attentions, is provably Bayes-optimal for an in-context recall task. When being trained with gradient descent, we show via a finite-sample analysis that the expected loss converges at linear rate to the Bayes risks. Moreover, we show that the trained transformers exhibit out-of-distribution (OOD) generalization, i.e., generalizing to samples outside of the population distribution. Our theoretical findings are further supported by extensive empirical validations, showing that \emph{without} proper parameterization, models with larger expressive power surprisingly \emph{fail} to generalize OOD after being trained by gradient descent.
Authors: Zeyuan Ma, Yue-Jiao Gong, Hongshu Guo, Wenjie Qiu, Sijie Ma, Hongqiao Lian, Jiajun Zhan, Kaixu Chen, Chen Wang, Zhiyang Huang, Zechuan Huang, Guojun Peng, Ran Cheng, Yining Ma
Abstract: Meta-Black-Box Optimization (MetaBBO) streamlines the automation of optimization algorithm design through meta-learning. It typically employs a bi-level structure: the meta-level policy undergoes meta-training to reduce the manual effort required in developing algorithms for low-level optimization tasks. The original MetaBox (2023) provided the first open-source framework for reinforcement learning-based single-objective MetaBBO. However, its relatively narrow scope no longer keep pace with the swift advancement in this field. In this paper, we introduce MetaBox-v2 (https://github.com/MetaEvo/MetaBox) as a milestone upgrade with four novel features: 1) a unified architecture supporting RL, evolutionary, and gradient-based approaches, by which we reproduce $23$ up-to-date baselines; 2) efficient parallelization schemes, which reduce the training/testing time by $10-40$x; 3) a comprehensive benchmark suite of $18$ synthetic/realistic tasks ($1900$+ instances) spanning single-objective, multi-objective, multi-model, and multi-task optimization scenarios; 4) plentiful and extensible interfaces for custom analysis/visualization and integrating to external optimization tools/benchmarks. To show the utility of MetaBox-v2, we carry out a systematic case study that evaluates the built-in baselines in terms of the optimization performance, generalization ability and learning efficiency. Valuable insights are concluded from thorough and detailed analysis for practitioners and those new to the field.
Authors: Mariano Rivera, Angello Hoyos
Abstract: We introduce CoLoRA (Convolutional Low-Rank Adaptation), a parameter-efficient fine-tuning method for convolutional neural networks (CNNs). CoLoRA extends LoRA to convolutional layers by decomposing kernel updates into lightweight depthwise and pointwise components.This design reduces the number of trainable parameters to 0.2 compared to conventional fine-tuning, preserves the original model size, and allows merging updates into the pretrained weights after each epoch, keeping inference complexity unchanged. On OCTMNISTv2, CoLoRA applied to VGG16 and ResNet50 achieves up to 1 percent accuracy and 0.013 AUC improvements over strong baselines (Vision Transformers, state-space, and Kolmogorov Arnold models) while reducing per-epoch training time by nearly 20 percent. Results indicate that CoLoRA provides a stable and effective alternative to full fine-tuning for medical image classification.
Authors: Pingbang Hu, Joseph Melkonian, Weijing Tang, Han Zhao, Jiaqi W. Ma
Abstract: Gradient-based data attribution methods, such as influence functions, are critical for understanding the impact of individual training samples without requiring repeated model retraining. However, their scalability is often limited by the high computational and memory costs associated with per-sample gradient computation. In this work, we propose GraSS, a novel gradient compression algorithm and its variants FactGraSS for linear layers specifically, that explicitly leverage the inherent sparsity of per-sample gradients to achieve sub-linear space and time complexity. Extensive experiments demonstrate the effectiveness of our approach, achieving substantial speedups while preserving data influence fidelity. In particular, FactGraSS achieves up to 165% faster throughput on billion-scale models compared to the previous state-of-the-art baselines. Our code is publicly available at https://github.com/TRAIS-Lab/GraSS.
Authors: Yang Xiao, Jiashuo Wang, Ruifeng Yuan, Chunpu Xu, Kaishuai Xu, Wenjie Li, Pengfei Liu
Abstract: Large language models (LLMs) have demonstrated remarkable reasoning capabilities through test-time scaling approaches, particularly when fine-tuned with chain-of-thought (CoT) data distilled from more powerful large reasoning models (LRMs). However, these reasoning chains often contain verbose elements that mirror human problem-solving, categorized as progressive reasoning (the essential solution development path) and functional elements (verification processes, alternative solution approaches, and error corrections). While progressive reasoning is crucial, the functional elements significantly increase computational demands during test-time inference. We introduce PIR (Perplexity-based Importance Refinement), a principled framework that quantitatively evaluates the importance of each reasoning step based on its impact on answer prediction confidence. PIR systematically identifies and selectively prunes only low-importance functional steps while preserving progressive reasoning components, creating optimized training data that maintains the integrity of the core solution path while reducing verbosity. Models fine-tuned on PIR-optimized data exhibit superior test-time scaling properties, generating more concise reasoning chains while achieving improved accuracy (+0.9\% to +6.6\%) with significantly reduced token usage (-3\% to -41\%) across challenging reasoning benchmarks (AIME, AMC, and GPQA Diamond). Our approach demonstrates strong generalizability across different model sizes, data sources, and token budgets, offering a practical solution for deploying reasoning-capable LLMs in scenarios where efficient test-time scaling, response time, and computational efficiency are valuable constraints.
Authors: Yufan Dang, Chen Qian, Xueheng Luo, Jingru Fan, Zihao Xie, Ruijie Shi, Weize Chen, Cheng Yang, Xiaoyin Che, Ye Tian, Xuantang Xiong, Lei Han, Zhiyuan Liu, Maosong Sun
Abstract: Large language models (LLMs) have achieved remarkable results across diverse downstream tasks, but their monolithic nature restricts scalability and efficiency in complex problem-solving. While recent research explores multi-agent collaboration among LLMs, most approaches rely on static organizational structures that struggle to adapt as task complexity and agent numbers grow, resulting in coordination overhead and inefficiencies. To this end, we propose a puppeteer-style paradigm for LLM-based multi-agent collaboration, where a centralized orchestrator ("puppeteer") dynamically directs agents ("puppets") in response to evolving task states. This orchestrator is trained via reinforcement learning to adaptively sequence and prioritize agents, enabling flexible and evolvable collective reasoning. Experiments on closed- and open-domain scenarios show that this method achieves superior performance with reduced computational costs. Analyses further reveal that the key improvements consistently stem from the emergence of more compact, cyclic reasoning structures under the orchestrator's evolution. Our code is available at https://github.com/OpenBMB/ChatDev/tree/puppeteer.
Authors: Zeinab Dehghani, Mohammed Naveed Akram, Koorosh Aslansefat, Adil Khan, Yiannis Papadopoulos
Abstract: Large Language Models (LLMs) such as GPT, LLaMA, and Claude achieve remarkable performance in text generation but remain opaque in their decision-making processes, limiting trust and accountability in high-stakes applications. We present gSMILE (generative SMILE), a model-agnostic, perturbation-based framework for token-level interpretability in LLMs. Extending the SMILE methodology, gSMILE uses controlled prompt perturbations, Wasserstein distance metrics, and weighted linear surrogates to identify input tokens with the most significant impact on the output. This process enables the generation of intuitive heatmaps that visually highlight influential tokens and reasoning paths. We evaluate gSMILE across leading LLMs (OpenAI's gpt-3.5-turbo-instruct, Meta's LLaMA 3.1 Instruct Turbo, and Anthropic's Claude 2.1) using attribution fidelity, attribution consistency, attribution stability, attribution faithfulness, and attribution accuracy as metrics. Results show that gSMILE delivers reliable human-aligned attributions, with Claude 2.1 excelling in attention fidelity and GPT-3.5 achieving the highest output consistency. These findings demonstrate gSMILE's ability to balance model performance and interpretability, enabling more transparent and trustworthy AI systems.
Authors: Yiran Guo, Lijie Xu, Jie Liu, Dan Ye, Shuang Qiu
Abstract: Enhancing the reasoning capabilities of large language models effectively using reinforcement learning (RL) remains a crucial challenge. Existing approaches primarily adopt two contrasting advantage estimation granularities: token-level methods (e.g., PPO) aim to provide fine-grained advantage signals but suffer from inaccurate estimation due to difficulties in training an accurate critic model. On the other extreme, trajectory-level methods (e.g., GRPO) solely rely on a coarse-grained advantage signal from the final reward, leading to imprecise credit assignment. To address these limitations, we propose Segment Policy Optimization (SPO), a novel RL framework that leverages segment-level advantage estimation at an intermediate granularity, achieving a better balance by offering more precise credit assignment than trajectory-level methods and requiring fewer estimation points than token-level methods, enabling accurate advantage estimation based on Monte Carlo (MC) without a critic model. SPO features three components with novel strategies: (1) flexible segment partition; (2) accurate segment advantage estimation; and (3) policy optimization using segment advantages, including a novel probability-mask strategy. We further instantiate SPO for two specific scenarios: (1) SPO-chain for short chain-of-thought (CoT), featuring novel cutpoint-based partition and chain-based advantage estimation, achieving $6$-$12$ percentage point improvements in accuracy over PPO and GRPO on GSM8K. (2) SPO-tree for long CoT, featuring novel tree-based advantage estimation, which significantly reduces the cost of MC estimation, achieving $7$-$11$ percentage point improvements over GRPO on MATH500 under 2K and 4K context evaluation. We make our code publicly available at https://github.com/AIFrameResearch/SPO.
Authors: Declan Kutscher, David M. Chan, Yutong Bai, Trevor Darrell, Ritwik Gupta
Abstract: Sequence models such as transformers require inputs to be represented as one-dimensional sequences. In vision, this typically involves flattening images using a fixed row-major (raster-scan) order. While full self-attention is permutation-equivariant, modern long-sequence transformers increasingly rely on architectural approximations that break this invariance and introduce sensitivity to patch ordering. We show that patch order significantly affects model performance in such settings, with simple alternatives like column-major or Hilbert curves yielding notable accuracy shifts. Motivated by this, we propose REOrder, a two-stage framework for discovering task-optimal patch orderings. First, we derive an information-theoretic prior by evaluating the compressibility of various patch sequences. Then, we learn a policy over permutations by optimizing a Plackett-Luce policy using REINFORCE. This approach enables efficient learning in a combinatorial permutation space. REOrder improves top-1 accuracy over row-major ordering on ImageNet-1K by up to 3.01% and Functional Map of the World by 13.35%.
Authors: Zhaokun Wang, Jinyu Guo, Jingwen Pu, Lingfeng Chen, Hongli Pu, Jie Ou, Libo Qin, Wenhong Tian
Abstract: Current parameter-efficient fine-tuning methods for adapting pre-trained language models to downstream tasks are susceptible to interference from noisy data. Conventional noise-handling approaches either rely on laborious data pre-processing or employ model architecture modifications prone to error accumulation. In contrast to existing noise-process paradigms, we propose a noise-robust adaptation method via asymmetric LoRA poisoning experts (LoPE), a novel framework that enhances model robustness to noise only with generated noisy data. Drawing inspiration from the mixture-of-experts architecture, LoPE strategically integrates a dedicated poisoning expert in an asymmetric LoRA configuration. Through a two-stage paradigm, LoPE performs noise injection on the poisoning expert during fine-tuning to enhance its noise discrimination and processing ability. During inference, we selectively mask the dedicated poisoning expert to leverage purified knowledge acquired by normal experts for noise-robust output. Extensive experiments demonstrate that LoPE achieves strong performance and robustness purely through the low-cost noise injection, which completely eliminates the requirement of data cleaning.
Authors: Hao Yan, Xingchen Liu, Hao Wang, Zhenbiao Cao, Handong Zheng, Liang Yin, Xinxing Su, Zihao Chen, Jihao Wu, Minghui Liao, Chao Weng, Wei Chen, Yuliang Liu, Xiang Bai
Abstract: Recent strides in multimodal large language models (MLLMs) have significantly advanced their performance in many reasoning tasks. However, Abstract Visual Reasoning (AVR) remains a critical challenge, primarily due to limitations in perceiving abstract graphics. To tackle this issue, we investigate the bottlenecks in current MLLMs and synthesize training data to improve their abstract visual perception. First, we propose VisuRiddles, a benchmark for AVR, featuring tasks meticulously constructed to assess models' reasoning capacities across five core dimensions and two high-level reasoning categories. Second, we introduce the Perceptual Riddle Synthesizer (PRS), an automated framework for generating riddles with fine-grained perceptual descriptions. PRS not only generates valuable training data for abstract graphics but also provides fine-grained perceptual description, crucially allowing for supervision over intermediate reasoning stages and thereby improving both training efficacy and model interpretability. Our extensive experimental results on VisuRiddles empirically validate that fine-grained visual perception is the principal bottleneck and our synthesis framework markedly enhances the performance of contemporary MLLMs on these challenging tasks. Our code and dataset will be released at https://github.com/yh-hust/VisuRiddles
Authors: Shihan Dou, Ming Zhang, Chenhao Huang, Jiayi Chen, Feng Chen, Shichun Liu, Yan Liu, Chenxiao Liu, Cheng Zhong, Zongzhang Zhang, Tao Gui, Chao Xin, Chengzhi Wei, Lin Yan, Yonghui Wu, Qi Zhang, Xuanjing Huang
Abstract: We introduce EvaLearn, a pioneering benchmark designed to evaluate large language models (LLMs) on their learning capability and efficiency in challenging tasks, a critical, yet underexplored aspect of model potential. EvaLearn contains 648 challenging problems across six task types, grouped into 182 sequences, each sequence dedicated to one task type. Diverging from most existing benchmarks that evaluate models in parallel, EvaLearn requires models to solve problems sequentially, allowing them to leverage the experience gained from previous solutions. EvaLearn provides five comprehensive automated metrics to evaluate models and quantify their learning capability and efficiency. We extensively benchmark nine frontier models and observe varied performance profiles: some models, such as Claude-3.7-sonnet, start with moderate initial performance but exhibit strong learning ability, while some models struggle to benefit from experience and may even show negative transfer. Moreover, we investigate model performance under two learning settings and find that instance-level rubrics and teacher-model feedback further facilitate model learning. Importantly, we observe that current LLMs with stronger static abilities do not show a clear advantage in learning capability across all tasks, highlighting that EvaLearn evaluates a new dimension of model performance. We hope EvaLearn provides a novel evaluation perspective for assessing LLM potential and understanding the gap between models and human capabilities, promoting the development of deeper and more dynamic evaluation approaches. All datasets, the automatic evaluation framework, and the results studied in this paper are available at the GitHub repository.
Authors: Baode Wang, Biao Wu, Weizhen Li, Meng Fang, Zuming Huang, Jun Huang, Haozhe Wang, Yanjie Liang, Ling Chen, Wei Chu, Yuan Qi
Abstract: Automated parsing of scanned documents into richly structured, machine-readable formats remains a critical bottleneck in Document AI, as traditional multi-stage pipelines suffer from error propagation and limited adaptability to diverse layouts. We introduce layoutRL, an end-to-end reinforcement learning framework that trains models to be explicitly layout-aware by optimizing a composite reward of normalized edit distance, paragraph count accuracy, and reading order preservation. Leveraging our newly released dataset, Infinity-Doc-55K, which combines 55K high-fidelity synthetic scanned document parsing data with expert-filtered real-world documents, we instantiate layoutRL in a vision-language-model-based parser called Infinity-Parser. Evaluated on English and Chinese benchmarks for OCR, table and formula extraction, and reading order detection, Infinity-Parser achieves new state-of-the-art performance in both accuracy and structural fidelity, outpacing specialist pipelines and general-purpose vision-language models. We will publicly release our code and dataset to accelerate progress in robust document understanding.
Authors: Moritz Miller, Bernhard Sch\"olkopf, Siyuan Guo
Abstract: Large-scale neural language models exhibit remarkable performance in in-context learning: the ability to learn and reason about the input context on the fly. This work studies in-context counterfactual reasoning in language models, that is, the ability to predict consequences of a hypothetical scenario. We focus on a well-defined, synthetic linear regression task that requires noise abduction. Accurate prediction is based on (1) inferring an unobserved latent concept and (2) copying contextual noise from factual observations. We show that language models are capable of counterfactual reasoning. Further, we enhance existing identifiability results and reduce counterfactual reasoning for a broad class of functions to a transformation on in-context observations. In Transformers, we find that self-attention, model depth and pre-training data diversity drive performance. Moreover, we provide mechanistic evidence that the latent concept is linearly represented in the residual stream and we introduce designated \textit{noise abduction heads} central to performing counterfactual reasoning. Lastly, our findings extend to counterfactual reasoning under SDE dynamics and reflect that Transformers can perform noise abduction on sequential data, providing preliminary evidence on the potential for counterfactual story generation. Our code is available under https://github.com/mrtzmllr/iccr.
Authors: Baptiste Chatelier (IETR, INSA Rennes, MERCE-France), Vincent Corlay (MERCE-France), Musa Furkan Keskin (INSA Rennes, IETR), Matthieu Crussi\`ere (INSA Rennes, IETR), Henk Wymeersch (INSA Rennes, IETR), Luc Le Magoarou (INSA Rennes, IETR)
Abstract: The increasing deployment of large antenna arrays at base stations has significantly improved the spatial resolution and localization accuracy of radio-localization methods. However, traditional signal processing techniques struggle in complex radio environments, particularly in scenarios dominated by non line of sight (NLoS) propagation paths, resulting in degraded localization accuracy. Recent developments in machine learning have facilitated the development of machine learning-assisted localization techniques, enhancing localization accuracy in complex radio environments. However, these methods often involve substantial computational complexity during both the training and inference phases. This work extends the well-established fingerprinting-based localization framework by simultaneously reducing its memory requirements and improving its accuracy. Specifically, a model-based neural network is used to learn the location-to-channel mapping, and then serves as a generative neural channel model. This generative model augments the fingerprinting comparison dictionary while reducing the memory requirements. The proposed method outperforms fingerprinting baselines by achieving sub-wavelength localization accuracy, even in complex static NLoS environments. Remarkably, it offers an improvement by several orders of magnitude in localization accuracy, while simultaneously reducing memory requirements by an order of magnitude compared to classical fingerprinting methods.
Authors: Jingyuan Ma, Rui Li, Zheng Li, Junfeng Liu, Heming Xia Lei Sha, Zhifang Sui
Abstract: Emerging Large Reasoning Models (LRMs) consistently excel in mathematical and reasoning tasks, showcasing remarkable capabilities. However, the enhancement of reasoning abilities and the exposure of internal reasoning processes introduce new safety vulnerabilities. A critical question arises: when reasoning becomes intertwined with harmfulness, will LRMs become more vulnerable to jailbreaks in reasoning mode? To investigate this, we introduce HauntAttack, a novel and general-purpose black-box adversarial attack framework that systematically embeds harmful instructions into reasoning questions. Specifically, we modify key reasoning conditions in existing questions with harmful instructions, thereby constructing a reasoning pathway that guides the model step by step toward unsafe outputs. We evaluate HauntAttack on 11 LRMs and observe an average attack success rate of 70\%, achieving up to 12 percentage points of absolute improvement over the strongest prior baseline. Our further analysis reveals that even advanced safety-aligned models remain highly susceptible to reasoning-based attacks, offering insights into the urgent challenge of balancing reasoning capability and safety in future model development.
Authors: Avishag Shapira, Parth Atulbhai Gandhi, Edan Habler, Asaf Shabtai
Abstract: Web-use agents are rapidly being deployed to automate complex web tasks with extensive browser capabilities. However, these capabilities create a critical and previously unexplored attack surface. This paper demonstrates how attackers can exploit web-use agents by embedding malicious content in web pages, such as comments, reviews, or advertisements, that agents encounter during legitimate browsing tasks. We introduce the task-aligned injection technique that frames malicious commands as helpful task guidance rather than obvious attacks, exploiting fundamental limitations in LLMs' contextual reasoning. Agents struggle to maintain coherent contextual awareness and fail to detect when seemingly helpful web content contains steering attempts that deviate them from their original task goal. To scale this attack, we developed an automated three-stage pipeline that generates effective injections without manual annotation or costly online agent interactions during training, remaining efficient even with limited training data. This pipeline produces a generator model that we evaluate on five popular agents using payloads organized by the Confidentiality-Integrity-Availability (CIA) security triad, including unauthorized camera activation, file exfiltration, user impersonation, phishing, and denial-of-service. This generator achieves over 80% attack success rate (ASR) with strong transferability across unseen payloads, diverse web environments, and different underlying LLMs. This attack succeed even against agents with built-in safety mechanisms, requiring only the ability to post content on public websites. To address this risk, we propose comprehensive mitigation strategies including oversight mechanisms, execution constraints, and task-aware reasoning techniques.
Authors: Florian Andreas Marwitz, Ralf M\"oller, Magnus Bender, Marcel Gehrke
Abstract: Inference in dynamic probabilistic models is a complex task involving expensive operations. In particular, for Hidden Markov Models, the whole state space has to be enumerated for advancing in time. Even states with negligible probabilities are considered, resulting in computational inefficiency and increased noise due to the propagation of unlikely probability mass. We propose to denoise the future and speed up inference by using only the top-p states, i.e., the most probable states with accumulated probability p. We show that the error introduced by using only the top-p states is bound by p and the so-called minimal mixing rate of the underlying model. Moreover, in our empirical evaluation, we show that we can expect speedups of at least an order of magnitude, while the error in terms of total variation distance is below 0.09.
Authors: Huaying Yuan, Zheng Liu, Junjie Zhou, Hongjin Qian, Yan Shu, Nicu Sebe, Ji-Rong Wen, Zhicheng Dou
Abstract: Long-video understanding~(LVU) is a challenging problem in computer vision. Existing methods either downsample frames for single-pass reasoning, sacrificing fine-grained details, or depend on textual reasoning over task-agnostic representations, hindering task-specific perception and exploration. In this paper, we propose VideoExplorer, a framework grounded in the principle of ``thinking with video'', which naturally intertwines planning, temporal grounding, and scalable perception into a coherent reasoning process. Rather than reasoning over a static context, VideoExplorer iteratively formulates sub-questions, locates relevant moments, and performs task-oriented, temporally scalable video understanding until reaching the final answer, enabling faithful, efficient, and interpretable reasoning. To address the lack of LVU training resources, we construct a long-video reasoning dataset using difficulty-adaptive sampling to ensure high-quality trajectories on complex tasks. Building on this dataset, we design a two-stage training pipeline: supervised trajectory initialization followed by trajectory-level preference optimization, encouraging adaptive temporal grounding and iterative information integration guided by downstream rewards. Extensive evaluations on popular long-video understanding and reasoning benchmarks demonstrate VideoExplorer's significant advantage over existing baselines, highlighting its robustness, adaptability, and efficiency. Our code is made publicly available in this repository(https://github.com/yhy-2000/VideoDeepResearch).
Authors: Zhaiming Shen, Alexander Hsu, Rongjie Lai, Wenjing Liao
Abstract: While in-context learning (ICL) has achieved remarkable success in natural language and vision domains, its theoretical understanding-particularly in the context of structured geometric data-remains unexplored. This paper initiates a theoretical study of ICL for regression of H\"older functions on manifolds. We establish a novel connection between the attention mechanism and classical kernel methods, demonstrating that transformers effectively perform kernel-based prediction at a new query through its interaction with the prompt. This connection is validated by numerical experiments, revealing that the learned query-prompt scores for H\"older functions are highly correlated with the Gaussian kernel. Building on this insight, we derive generalization error bounds in terms of the prompt length and the number of training tasks. When a sufficient number of training tasks are observed, transformers give rise to the minimax regression rate of H\"older functions on manifolds, which scales exponentially with the intrinsic dimension of the manifold, rather than the ambient space dimension. Our result also characterizes how the generalization error scales with the number of training tasks, shedding light on the complexity of transformers as in-context kernel algorithm learners. Our findings provide foundational insights into the role of geometry in ICL and novels tools to study ICL of nonlinear models.
Authors: Haritz Puerto, Martin Gubri, Tommaso Green, Seong Joon Oh, Sangdoo Yun
Abstract: Large Language Models (LLMs) are transforming search engines into Conversational Search Engines (CSE). Consequently, Search Engine Optimization (SEO) is being shifted into Conversational Search Engine Optimization (C-SEO). We are beginning to see dedicated C-SEO methods for modifying web documents to increase their visibility in CSE responses. However, they are often tested only for a limited breadth of application domains; we do not know whether certain C-SEO methods would be effective for a broad range of domains. Moreover, existing evaluations consider only a single-actor scenario where only one web document adopts a C-SEO method; in reality, multiple players are likely to competitively adopt the cutting-edge C-SEO techniques, drawing an analogy from the dynamics we have seen in SEO. We present C-SEO Bench, the first benchmark designed to evaluate C-SEO methods across multiple tasks, domains, and number of actors. We consider two search tasks, question answering and product recommendation, with three domains each. We also formalize a new evaluation protocol with varying adoption rates among involved actors. Our experiments reveal that most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking, which is opposite to what is expected. Instead, traditional SEO strategies, those aiming to improve the ranking of the source in the LLM context, are significantly more effective. We also observe that as we increase the number of C-SEO adopters, the overall gains decrease, depicting a congested and zero-sum nature of the problem. Our code and data are available at https://github.com/parameterlab/c-seo-bench and https://huggingface.co/datasets/parameterlab/c-seo-bench.
URLs: https://github.com/parameterlab/c-seo-bench, https://huggingface.co/datasets/parameterlab/c-seo-bench.
Authors: Ruixuan Deng, Xiaoyang Hu, Miles Gilberti, Shane Storks, Aman Taxali, Mike Angstadt, Chandra Sripada, Joyce Chai
Abstract: We identify semantically coherent, context-consistent network components in large language models (LLMs) using coactivation of sparse autoencoder (SAE) features collected from just a handful of prompts. Focusing on concept-relation prediction tasks, we show that ablating these components for concepts (e.g., countries and words) and relations (e.g., capital city and translation language) changes model outputs in predictable ways, while amplifying these components induces counterfactual responses. Notably, composing relation and concept components yields compound counterfactual outputs. Further analysis reveals that while most concept components emerge from the very first layer, more abstract relation components are concentrated in later layers. Lastly, we show that extracted components more comprehensively capture concepts and relations than individual features while maintaining specificity. Overall, our findings suggest a modular organization of knowledge accessed through compositional operations, and advance methods for efficient, targeted LLM manipulation.
Authors: Nasa Matsumoto, Quoc Hoan Tran, Koki Chinzei, Yasuhiro Endo, Hirotaka Oshima
Abstract: Quantum machine learning models that leverage quantum circuits as quantum feature maps (QFMs) are recognized for their enhanced expressive power in learning tasks. Such models have demonstrated rigorous end-to-end quantum speedups for specific families of classification problems. However, deploying deep QFMs on real quantum hardware remains challenging due to circuit noise and hardware constraints. Additionally, variational quantum algorithms often suffer from computational bottlenecks, particularly in accurate gradient estimation, which significantly increases quantum resource demands during training. We propose Iterative Quantum Feature Maps (IQFMs), a hybrid quantum-classical framework that constructs a deep architecture by iteratively connecting shallow QFMs with classically computed augmentation weights. By incorporating contrastive learning and a layer-wise training mechanism, the IQFMs framework effectively reduces quantum runtime and mitigates noise-induced degradation. In tasks involving noisy quantum data, numerical experiments show that the IQFMs framework outperforms quantum convolutional neural networks, without requiring the optimization of variational quantum parameters. Even for a typical classical image classification benchmark, a carefully designed IQFMs framework achieves performance comparable to that of classical neural networks. This framework presents a promising path to address current limitations and harness the full potential of quantum-enhanced machine learning.
Authors: Hao Liu, Yu Hu, Rakiba Rayhana, Ling Bai, Zheng Liu
Abstract: Bed-related falls remain a major source of injury in hospitals and long-term care facilities, yet many commercial alarms trigger only after a patient has already left the bed. We show that early bed-exit intent can be predicted using only one low-cost load cell mounted under a bed leg. The resulting load signals are first converted into a compact set of complementary images: an RGB line plot that preserves raw waveforms and three texture maps-recurrence plot, Markov transition field, and Gramian angular field-that expose higher-order dynamics. We introduce ViFusionTST, a dual-stream Swin Transformer that processes the line plot and texture maps in parallel and fuses them through cross-attention to learn data-driven modality weights. To provide a realistic benchmark, we collected six months of continuous data from 95 beds in a long-term-care facility. On this real-world dataset ViFusionTST reaches an accuracy of 0.885 and an F1 score of 0.794, surpassing recent 1D and 2D time-series baselines across F1, recall, accuracy, and AUPRC. The results demonstrate that image-based fusion of load-sensor signals for time series classification is a practical and effective solution for real-time, privacy-preserving fall prevention.
Authors: Niket Patel, Randall Balestriero
Abstract: The grand goal of AI research, and particularly Self Supervised Learning (SSL), is to produce systems that can successfully solve any possible task. In contrast, current evaluation methods available to AI researchers typically rely on a fixed collection of hand-picked downstream benchmarks. Hence, a large amount of effort is put into designing and searching for large collection of evaluation tasks that can serve as a proxy of our grand goal. We argue that such a rigid evaluation protocol creates a silent bottleneck in AI research. To remedy that, we define a probabilistic space of downstream tasks obtained by adopting a distribution of tasks and by defining Task Priors. Under this view, one can evaluate a model's performance over the set of all possible downstream tasks. Our framework is the first to provide answers to key questions such as (i) what is the average performance of my model over all possible downstream tasks weighted by the probability to encounter each task? or (ii) what is the variance of my model's performance across all downstream tasks under the defined Task Priors? Beyond establishing a new standard for evaluation, we believe that Task Priors will accelerate the pace of research in SSL - where downstream task evaluation is the sole qualitative signal that researchers have access to.
Authors: Tim Tian Hua, James Baskerville, Henri Lemoine, Mia Hopman, Aryan Bhatt, Tyler Tracy
Abstract: Monitoring AIs at runtime can help us detect and stop harmful actions. In this paper, we study how to efficiently combine multiple runtime monitors into a single monitoring protocol. The protocol's objective is to maximize the probability of applying a safety intervention on misaligned outputs (i.e., maximize recall). Since running monitors and applying safety interventions are costly, the protocol also needs to adhere to an average-case budget constraint. Taking the monitors' performance and cost as given, we develop an algorithm to find the best protocol. The algorithm exhaustively searches over when and which monitors to call, and allocates safety interventions based on the Neyman-Pearson lemma. By focusing on likelihood ratios and strategically trading off spending on monitors against spending on interventions, we more than double our recall rate compared to a naive baseline in a code review setting. We also show that combining two monitors can Pareto dominate using either monitor alone. Our framework provides a principled methodology for combining existing monitors to detect undesirable behavior in cost-sensitive settings.
Authors: Wenbin Guo, Xin Wang, Jiaoyan Chen, Zhao Li, Zirui Chen
Abstract: Large Language Models (LLMs) have been extensively adopted in Knowledge Graph Completion (KGC), showcasing significant research advancements. However, as black-box models driven by deep neural architectures, current LLM-based KGC methods rely on implicit knowledge representation with parallel propagation of erroneous knowledge, thereby hindering their ability to produce conclusive and decisive reasoning outcomes. We aim to integrate neural-perceptual structural information with ontological knowledge, leveraging the powerful capabilities of LLMs to achieve a deeper understanding of the intrinsic logic of the knowledge. We propose an ontology enhanced KGC method using LLMs -- OL-KGC. It first leverages neural perceptual mechanisms to effectively embed structural information into the textual space, and then uses an automated extraction algorithm to retrieve ontological knowledge from the knowledge graphs (KGs) that needs to be completed, which is further transformed into a textual format comprehensible to LLMs for providing logic guidance. We conducted extensive experiments on three widely-used benchmarks -- FB15K-237, UMLS and WN18RR. The experimental results demonstrate that OL-KGC significantly outperforms existing mainstream KGC methods across multiple evaluation metrics, achieving state-of-the-art performance.
Authors: Matteo Giacomini, Antonio Huerta
Abstract: A surrogate-based topology optimisation algorithm for linear elastic structures under parametric loads and boundary conditions is proposed. Instead of learning the parametric solution of the state (and adjoint) problems or the optimisation trajectory as a function of the iterations, the proposed approach devises a surrogate version of the entire optimisation pipeline. First, the method predicts a quasi-optimal topology for a given problem configuration as a surrogate model of high-fidelity topologies optimised with the homogenisation method. This is achieved by means of a feed-forward net learning the mapping between the input parameters characterising the system setup and a latent space determined by encoder/decoder blocks reducing the dimensionality of the parametric topology optimisation problem and reconstructing a high-dimensional representation of the topology. Then, the predicted topology is used as an educated initial guess for a computationally efficient algorithm penalising the intermediate values of the design variable, while enforcing the governing equations of the system. This step allows the method to correct potential errors introduced by the surrogate model, eliminate artifacts, and refine the design in order to produce topologies consistent with the underlying physics. Different architectures are proposed and the approximation and generalisation capabilities of the resulting models are numerically evaluated. The quasi-optimal topologies allow to outperform the high-fidelity optimiser by reducing the average number of optimisation iterations by $53\%$ while achieving discrepancies below $4\%$ in the optimal value of the objective functional, even in the challenging scenario of testing the model to extrapolate beyond the training and validation domain.
Authors: Linghao Zhu, Yiran Guan, Dingkang Liang, Jianzhong Ju, Zhenbo Luo, Bin Qin, Jian Luan, Yuliang Liu, Xiang Bai
Abstract: Reinforcement learning (RL) has emerged as an effective post-training paradigm for enhancing the reasoning capabilities of multimodal large language model (MLLM). However, current RL pipelines often suffer from training inefficiencies caused by two underexplored issues: Advantage Collapsing, where most advantages in a batch concentrate near zero, and Rollout Silencing, where the proportion of rollouts contributing non-zero gradients diminishes over time. These issues lead to suboptimal gradient updates and hinder long-term learning efficiency. To address these issues, we propose Shuffle-R1, a simple yet principled framework that improves RL fine-tuning efficiency by dynamically restructuring trajectory sampling and batch composition. It introduces (1) Pairwise Trajectory Sampling, which selects high-contrast trajectories with large advantages to improve gradient signal quality, and (2) Advantage-based Trajectory Shuffle, which increases exposure of valuable rollouts through informed batch reshuffling. Experiments across multiple reasoning benchmarks show that our framework consistently outperforms strong RL baselines with minimal overhead. These results highlight the importance of data-centric adaptations for more efficient RL training in MLLM.
Authors: Shuang Liang, Zhihao Xu, Jialing Tao, Hui Xue, Xiting Wang
Abstract: Despite extensive alignment efforts, Large Vision-Language Models (LVLMs) remain vulnerable to jailbreak attacks, posing serious safety risks. To address this, existing detection methods either learn attack-specific parameters, which hinders generalization to unseen attacks, or rely on heuristically sound principles, which limit accuracy and efficiency. To overcome these limitations, we propose Learning to Detect (LoD), a general framework that accurately detects unknown jailbreak attacks by shifting the focus from attack-specific learning to task-specific learning. This framework includes a Multi-modal Safety Concept Activation Vector module for safety-oriented representation learning and a Safety Pattern Auto-Encoder module for unsupervised attack classification. Extensive experiments show that our method achieves consistently higher detection AUROC on diverse unknown attacks while improving efficiency. The code is available at https://anonymous.4open.science/r/Learning-to-Detect-51CB.
URLs: https://anonymous.4open.science/r/Learning-to-Detect-51CB.
Authors: Andrej Orsula, Matthieu Geist, Miguel Olivares-Mendez, Carol Martinez
Abstract: Reliable autonomous navigation across the unstructured terrains of distant planetary surfaces is a critical enabler for future space exploration. However, the deployment of learning-based controllers is hindered by the inherent sim-to-real gap, particularly for the complex dynamics of wheel interactions with granular media. This work presents a complete sim-to-real framework for developing and validating robust control policies for dynamic waypoint tracking on such challenging surfaces. We leverage massively parallel simulation to train reinforcement learning agents across a vast distribution of procedurally generated environments with randomized physics. These policies are then transferred zero-shot to a physical wheeled rover operating in a lunar-analogue facility. Our experiments systematically compare multiple reinforcement learning algorithms and action smoothing filters to identify the most effective combinations for real-world deployment. Crucially, we provide strong empirical evidence that agents trained with procedural diversity achieve superior zero-shot performance compared to those trained on static scenarios. We also analyze the trade-offs of fine-tuning with high-fidelity particle physics, which offers minor gains in low-speed precision at a significant computational cost. Together, these contributions establish a validated workflow for creating reliable learning-based navigation systems, marking a substantial step towards deploying autonomous robots in the final frontier.
Authors: Jonas van Elburg, Peter van der Putten, Maarten Marx
Abstract: We investigate whether synthetic question-answer (QA) data generated by large language models (LLMs) can serve as an effective proxy for human-labeled benchmarks when the latter is unavailable. We assess the reliability of synthetic benchmarks across two experiments: one varying retriever parameters while keeping the generator fixed, and another varying the generator with fixed retriever parameters. Across four datasets, of which two open-domain and two proprietary, we find that synthetic benchmarks reliably rank the RAGs varying in terms of retriever configuration, aligning well with human-labeled benchmark baselines. However, they do not consistently produce reliable RAG rankings when comparing generator architectures. The breakdown possibly arises from a combination of task mismatch between the synthetic and human benchmarks, and stylistic bias favoring certain generators.
Authors: Mona Mirzaie, Bodo Rosenhahn
Abstract: Trustworthy AI is mandatory for the broad deployment of autonomous vehicles. Although end-to-end approaches derive control commands directly from raw data, interpreting these decisions remains challenging, especially in complex urban scenarios. This is mainly attributed to very deep neural networks with non-linear decision boundaries, making it challenging to grasp the logic behind AI-driven decisions. This paper presents a method to enhance interpretability while optimizing control commands in autonomous driving. To address this, we propose loss functions that promote the interpretability of our model by generating sparse and localized feature maps. The feature activations allow us to explain which image regions contribute to the predicted control command. We conduct comprehensive ablation studies on the feature extraction step and validate our method on the CARLA benchmarks. We also demonstrate that our approach improves interpretability, which correlates with reducing infractions, yielding a safer, high-performance driving model. Notably, our monocular, non-ensemble model surpasses the top-performing approaches from the CARLA Leaderboard by achieving lower infraction scores and the highest route completion rate, all while ensuring interpretability.
Authors: Amine Lbath, Massih-Reza Amini, Aurelien Delaitre, Vadim Okun
Abstract: The increasing complexity of software systems and the sophistication of cyber-attacks have underscored the critical need for effective automated vulnerability detection and repair systems. Data-driven approaches using deep learning models show promise but critically depend on the availability of large, accurately labeled datasets. Yet existing datasets either suffer from noisy labels, limited range of vulnerabilities, or fail to reflect vulnerabilities as they occur in real-world software. This also limits large-scale benchmarking of such solutions. Automated vulnerability injection provides a way to directly address these dataset limitations, but existing techniques remain limited in coverage, contextual fidelity, or injection success rates. In this paper, we present AVIATOR, the first AI-agentic vulnerability injection workflow. It automatically injects realistic, category-specific vulnerabilities for high-fidelity, diverse, large-scale vulnerability dataset generation. Unlike prior monolithic approaches, AVIATOR orchestrates specialized AI agents, function agents and traditional code analysis tools that replicate expert reasoning. It combines semantic analysis, injection synthesis enhanced with LoRA-based fine-tuning and Retrieval-Augmented Generation, as well as post-injection validation via static analysis and LLM-based discriminators. This modular decomposition allows specialized agents to focus on distinct tasks, improving robustness of injection and reducing error propagation across the workflow. Evaluations across three distinct benchmarks demonstrate that AVIATOR achieves 91%-95% injection success rates, significantly surpassing existing automated dataset generation techniques in both accuracy and scope of software vulnerabilities.
Authors: Cheonsu Jeong, Seunghyun Lee, Seonhee Jeong, Sungsu Kim
Abstract: This study provides an in_depth analysis of the ethical and trustworthiness challenges emerging alongside the rapid advancement of generative artificial intelligence (AI) technologies and proposes a comprehensive framework for their systematic evaluation. While generative AI, such as ChatGPT, demonstrates remarkable innovative potential, it simultaneously raises ethical and social concerns, including bias, harmfulness, copyright infringement, privacy violations, and hallucination. Current AI evaluation methodologies, which mainly focus on performance and accuracy, are insufficient to address these multifaceted issues. Thus, this study emphasizes the need for new human_centered criteria that also reflect social impact. To this end, it identifies key dimensions for evaluating the ethics and trustworthiness of generative AI_fairness, transparency, accountability, safety, privacy, accuracy, consistency, robustness, explainability, copyright and intellectual property protection, and source traceability and develops detailed indicators and assessment methodologies for each. Moreover, it provides a comparative analysis of AI ethics policies and guidelines in South Korea, the United States, the European Union, and China, deriving key approaches and implications from each. The proposed framework applies across the AI lifecycle and integrates technical assessments with multidisciplinary perspectives, thereby offering practical means to identify and manage ethical risks in real_world contexts. Ultimately, the study establishes an academic foundation for the responsible advancement of generative AI and delivers actionable insights for policymakers, developers, users, and other stakeholders, supporting the positive societal contributions of AI technologies.
Authors: Fangzhou Wu, Sandeep Silwal
Abstract: Increasing demand for Large Language Models (LLMs) services imposes substantial deployment and computation costs on providers. LLM routing offers a cost-efficient solution by directing queries to the optimal LLM based on model and query features. However, existing works primarily focus on offline scenarios and struggle to adapt to online settings with high query volume and constrained token budgets. In this work, we introduce the first training-free algorithm for online routing scenarios. Our algorithm leverages approximate nearest neighbor search to efficiently estimate query features and performs a one-time optimization over a small set of initial queries to learn a routing strategy that guides future routing. We provide theoretical guarantees demonstrating that our algorithm achieves a competitive ratio of $1 - o(1)$ under natural assumptions, which is further validated by extensive experiments across 3 benchmark datasets and 8 baselines, showing an average improvement of 3.55$\times$ in overall performance, 1.85$\times$ in cost efficiency, and nearly 4.25$\times$ in throughput. Our code is available at https://github.com/fzwark/PORT.
Authors: Rohit Patel
Abstract: This paper provides a self-contained, from-scratch, exposition of key algorithms for instruction tuning of models: SFT, Rejection Sampling, REINFORCE, Trust Region Policy Optimization (TRPO), Proximal Policy Optimization (PPO), Group Relative Policy Optimization (GRPO), and Direct Preference Optimization (DPO). Explanations of these algorithms often assume prior knowledge, lack critical details, and/or are overly generalized and complex. Here, each method is discussed and developed step by step using simplified and explicit notation focused on LLMs, aiming to eliminate ambiguity and provide a clear and intuitive understanding of the concepts. By minimizing detours into the broader RL literature and connecting concepts to LLMs, we eliminate superfluous abstractions and reduce cognitive overhead. Following this exposition, we provide a literature review of new techniques and approaches beyond those detailed. Finally, new ideas for research and exploration in the form of GRAPE (Generalized Relative Advantage Policy Evolution) are presented.
Authors: Jialin Wu, Shreya Saha, Yiqing Bo, Meenakshi Khosla
Abstract: Representational similarity metrics are fundamental tools in neuroscience and AI, yet we lack systematic comparisons of their discriminative power across model families. We introduce a quantitative framework to evaluate representational similarity measures based on their ability to separate model families-across architectures (CNNs, Vision Transformers, Swin Transformers, ConvNeXt) and training regimes (supervised vs. self-supervised). Using three complementary separability measures-dprime from signal detection theory, silhouette coefficients and ROC-AUC, we systematically assess the discriminative capacity of commonly used metrics including RSA, linear predictivity, Procrustes, and soft matching. We show that separability systematically increases as metrics impose more stringent alignment constraints. Among mapping-based approaches, soft-matching achieves the highest separability, followed by Procrustes alignment and linear predictivity. Non-fitting methods such as RSA also yield strong separability across families. These results provide the first systematic comparison of similarity metrics through a separability lens, clarifying their relative sensitivity and guiding metric choice for large-scale model and brain comparisons.
Authors: Jie Zhang, Ting Xu, Gelei Deng, Runyi Hu, Han Qiu, Tianwei Zhang, Qing Guo, Ivor Tsang
Abstract: Writing is a universal cultural technology that reuses vision for symbolic communication. Humans display striking resilience: we readily recognize words even when characters are fragmented, fused, or partially occluded. This paper investigates whether advanced vision language models (VLMs) share this resilience. We construct two psychophysics inspired benchmarks across distinct writing systems, Chinese logographs and English alphabetic words, by splicing, recombining, and overlaying glyphs to yield ''visible but unreadable'' stimuli for models while remaining legible to humans. Despite strong performance on clean text, contemporary VLMs show a severe drop under these perturbations, frequently producing unrelated or incoherent outputs. The pattern suggests a structural limitation: models heavily leverage generic visual invariances but under rely on compositional priors needed for robust literacy. We release stimuli generation code, prompts, and evaluation protocols to facilitate transparent replication and follow up work. Our findings motivate architectures and training strategies that encode symbol segmentation, composition, and binding across scripts, and they delineate concrete challenges for deploying multimodal systems in education, accessibility, cultural heritage, and security.
Authors: Sepehr Golrokh Amin, Devin Rhoads, Fatemeh Fakhrmoosavi, Nicholas E. Lownes, John N. Ivan
Abstract: This study introduces a Large Language Model (LLM) scheme for generating individual travel diaries in agent-based transportation models. While traditional approaches rely on large quantities of proprietary household travel surveys, the method presented in this study generates personas stochastically from open-source American Community Survey (ACS) and Smart Location Database (SLD) data, then synthesizes diaries through direct prompting. This study features a novel one-to-cohort realism score: a composite of four metrics (Trip Count Score, Interval Score, Purpose Score, and Mode Score) validated against the Connecticut Statewide Transportation Study (CSTS) diaries, matched across demographic variables. The validation utilizes Jensen-Shannon Divergence to measure distributional similarities between generated and real diaries. When compared to diaries generated with classical methods (Negative Binomial for trip generation; Multinomial Logit for mode/purpose) calibrated on the validation set, LLM-generated diaries achieve comparable overall realism (LLM mean: 0.485 vs. 0.455). The LLM excels in determining trip purpose and demonstrates greater consistency (narrower realism score distribution), while classical models lead in numerical estimates of trip count and activity duration. Aggregate validation confirms the LLM's statistical representativeness (LLM mean: 0.612 vs. 0.435), demonstrating LLM's zero-shot viability and establishing a quantifiable metric of diary realism for future synthetic diary evaluation systems.
Authors: Amber Shore, Russell Scheinberg, Ameeta Agrawal, So Young Lee
Abstract: Large Language Models (LLMs) are intended to reflect human linguistic competencies. But humans have access to a broad and embodied context, which is key in detecting and resolving linguistic ambiguities, even in isolated text spans. A foundational case of semantic ambiguity is found in the task of coreference resolution: how is a pronoun related to an earlier person mention? This capability is implicit in nearly every downstream task, and the presence of ambiguity at this level can alter performance significantly. We show that LLMs can achieve good performance with minimal prompting in both coreference disambiguation and the detection of ambiguity in coreference, however, they cannot do both at the same time. We present the CORRECT-DETECT trade-off: though models have both capabilities and deploy them implicitly, successful performance balancing these two abilities remains elusive.
Authors: Amirhossein Yousefiramandi, Ciaran Cooney
Abstract: Transformer-based language models such as BERT have become foundational in NLP, yet their performance degrades in specialized domains like patents, which contain long, technical, and legally structured text. Prior approaches to patent NLP have primarily relied on fine-tuning general-purpose models or domain-adapted variants pretrained with limited data. In this work, we pretrain 3 domain-specific masked language models for patents, using the ModernBERT architecture and a curated corpus of over 60 million patent records. Our approach incorporates architectural optimizations, including FlashAttention, rotary embeddings, and GLU feed-forward layers. We evaluate our models on four downstream patent classification tasks. Our model, ModernBERT-base-PT, consistently outperforms the general-purpose ModernBERT baseline on three out of four datasets and achieves competitive performance with a baseline PatentBERT. Additional experiments with ModernBERT-base-VX and Mosaic-BERT-large demonstrate that scaling the model size and customizing the tokenizer further enhance performance on selected tasks. Notably, all ModernBERT variants retain substantially faster inference over - 3x that of PatentBERT - underscoring their suitability for time-sensitive applications. These results underscore the benefits of domain-specific pretraining and architectural improvements for patent-focused NLP tasks.
Authors: Riccardo Cadei, Christian Intern\`o
Abstract: Modern foundational models increasingly reflect not just world knowledge, but patterns of human preference embedded in their training data. We hypothesize that recursive alignment-via human feedback and model-generated corpora-induces a social desirability bias, nudging models to favor agreeable or flattering responses over objective reasoning. We refer to it as the Narcissus Hypothesis and test it across 31 models using standardized personality assessments and a novel Social Desirability Bias score. Results reveal a significant drift toward socially conforming traits, with profound implications for corpus integrity and the reliability of downstream inferences. We then offer a novel epistemological interpretation, tracing how recursive bias may collapse higher-order reasoning down Pearl's Ladder of Causality, culminating in what we refer to as the Rung of Illusion.
Authors: Ye Liu, Zongyang Ma, Junfu Pu, Zhongang Qi, Yang Wu, Ying Shan, Chang Wen Chen
Abstract: Recent advances in Large Multi-modal Models (LMMs) have demonstrated their remarkable success as general-purpose multi-modal assistants, with particular focuses on holistic image- and video-language understanding. Conversely, less attention has been given to scaling fine-grained pixel-level understanding capabilities, where the models are expected to realize pixel-level alignment between visual signals and language semantics. Some previous studies have applied LMMs to related tasks such as region-level captioning and referring expression segmentation. However, these models are limited to performing either referring or segmentation tasks independently and fail to integrate these fine-grained perception capabilities into visual reasoning. To bridge this gap, we propose UniPixel, a large multi-modal model capable of flexibly comprehending visual prompt inputs and generating mask-grounded responses. Our model distinguishes itself by seamlessly integrating pixel-level perception with general visual understanding capabilities. Specifically, UniPixel processes visual prompts and generates relevant masks on demand, and performs subsequent reasoning conditioning on these intermediate pointers during inference, thereby enabling fine-grained pixel-level reasoning. The effectiveness of our approach has been verified on 10 benchmarks across a diverse set of tasks, including pixel-level referring/segmentation and object-centric understanding in images/videos. A novel PixelQA task that jointly requires referring, segmentation, and question answering is also designed to verify the flexibility of our method.
Authors: Zhenyu Tao, Wei Xu, Xiaohu You
Abstract: The bisimulation metric (BSM) is a powerful tool for computing state similarities within a Markov decision process (MDP), revealing that states closer in BSM have more similar optimal value functions. While BSM has been successfully utilized in reinforcement learning (RL) for tasks like state representation learning and policy exploration, its application to multiple-MDP scenarios, such as policy transfer, remains challenging. Prior work has attempted to generalize BSM to pairs of MDPs, but a lack of rigorous analysis of its mathematical properties has limited further theoretical progress. In this work, we formally establish a generalized bisimulation metric (GBSM) between pairs of MDPs, which is rigorously proven with the three fundamental properties: GBSM symmetry, inter-MDP triangle inequality, and the distance bound on identical state spaces. Leveraging these properties, we theoretically analyse policy transfer, state aggregation, and sampling-based estimation in MDPs, obtaining explicit bounds that are strictly tighter than those derived from the standard BSM. Additionally, GBSM provides a closed-form sample complexity for estimation, improving upon existing asymptotic results based on BSM. Numerical results validate our theoretical findings and demonstrate the effectiveness of GBSM in multi-MDP scenarios.
Authors: Haochen Gong, Chenxiao Li, Rui Chang, Wenbo Shen
Abstract: Large language model (LLM)-based computer-use agents represent a convergence of AI and OS capabilities, enabling natural language to control system- and application-level functions. However, due to LLMs' inherent uncertainty issues, granting agents control over computers poses significant security risks. When agent actions deviate from user intentions, they can cause irreversible consequences. Existing mitigation approaches, such as user confirmation and LLM-based dynamic action validation, still suffer from limitations in usability, security, and performance. To address these challenges, we propose CSAgent, a system-level, static policy-based access control framework for computer-use agents. To bridge the gap between static policy and dynamic context and user intent, CSAgent introduces intent- and context-aware policies, and provides an automated toolchain to assist developers in constructing and refining them. CSAgent enforces these policies through an optimized OS service, ensuring that agent actions can only be executed under specific user intents and contexts. CSAgent supports protecting agents that control computers through diverse interfaces, including API, CLI, and GUI. We implement and evaluate CSAgent, which successfully defends against more than 99.36% of attacks while introducing only 6.83% performance overhead.
Authors: Nimisha Ghosh, Dheeran Sankaran, Rahul Balakrishnan Adhi, Sharath S, Amrut Anand
Abstract: Identifying DNA- (DBPs) and RNA-binding proteins (RBPs) is crucial for the understanding of cell function, molecular interactions as well as regulatory functions. Owing to their high similarity, most of the existing approaches face challenges in differentiating between DBPs and RBPs leading to high cross-prediction errors. Moreover, identifying proteins which bind to both DNA and RNA (DRBPs) is also quite a challenging task. In this regard, we propose a novel framework viz. LAMP-PRo which is based on pre-trained protein language model (PLM), attention mechanisms and multi-label learning to mitigate these issues. First, pre-trained PLM such ESM-2 is used for embedding the protein sequences followed by convolutional neural network (CNN). Subsequently multi-head self-attention mechanism is applied for the contextual information while label-aware attention is used to compute class-specific representations by attending to the sequence in a way that is tailored to each label (DBP, RBP and non-NABP) in a multi-label setup. We have also included a novel cross-label attention mechanism to explicitly capture dependencies between DNA- and RNA-binding proteins, enabling more accurate prediction of DRBP. Finally, a linear layer followed by a sigmoid function are used for the final prediction. Extensive experiments are carried out to compare LAMP-PRo with the existing methods wherein the proposed model shows consistent competent performance. Furthermore, we also provide visualization to showcase model interpretability, highlighting which parts of the sequence are most relevant for a predicted label. The original datasets are available at http://bliulab.net/iDRBP\_MMC and the codes are available at https://github.com/NimishaGhosh/LAMP-PRo.
URLs: http://bliulab.net/iDRBP\_MMC, https://github.com/NimishaGhosh/LAMP-PRo.
Authors: Pranav Saxena, Avigyan Bhattacharya, Ji Zhang, Wenshan Wang
Abstract: Referential grounding in outdoor driving scenes is challenging due to large scene variability, many visually similar objects, and dynamic elements that complicate resolving natural-language references (e.g., "the black car on the right"). We propose LLM-RG, a hybrid pipeline that combines off-the-shelf vision-language models for fine-grained attribute extraction with large language models for symbolic reasoning. LLM-RG processes an image and a free-form referring expression by using an LLM to extract relevant object types and attributes, detecting candidate regions, generating rich visual descriptors with a VLM, and then combining these descriptors with spatial metadata into natural-language prompts that are input to an LLM for chain-of-thought reasoning to identify the referent's bounding box. Evaluated on the Talk2Car benchmark, LLM-RG yields substantial gains over both LLM and VLM-based baselines. Additionally, our ablations show that adding 3D spatial cues further improves grounding. Our results demonstrate the complementary strengths of VLMs and LLMs, applied in a zero-shot manner, for robust outdoor referential grounding.
Authors: Jinghang Shi, Xiaoyu Tang, Yang Huang, Yuyang Li, Xiao Kong, Yanxia Zhang, Caizhan Yue
Abstract: Astronomical image interpretation presents a significant challenge for applying multimodal large language models (MLLMs) to specialized scientific tasks. Existing benchmarks focus on general multimodal capabilities but fail to capture the complexity of astronomical data. To bridge this gap, we introduce AstroMMBench, the first comprehensive benchmark designed to evaluate MLLMs in astronomical image understanding. AstroMMBench comprises 621 multiple-choice questions across six astrophysical subfields, curated and reviewed by 15 domain experts for quality and relevance. We conducted an extensive evaluation of 25 diverse MLLMs, including 22 open-source and 3 closed-source models, using AstroMMBench. The results show that Ovis2-34B achieved the highest overall accuracy (70.5%), demonstrating leading capabilities even compared to strong closed-source models. Performance showed variations across the six astrophysical subfields, proving particularly challenging in domains like cosmology and high-energy astrophysics, while models performed relatively better in others, such as instrumentation and solar astrophysics. These findings underscore the vital role of domain-specific benchmarks like AstroMMBench in critically evaluating MLLM performance and guiding their targeted development for scientific applications. AstroMMBench provides a foundational resource and a dynamic tool to catalyze advancements at the intersection of AI and astronomy.
Authors: Aadarsh Rajiv Patel, Klaus Mueller
Abstract: Modern legislative frameworks, such as the Affordable Care Act (ACA), often involve complex webs of agencies, mandates, and interdependencies. Government issued charts attempt to depict these structures but are typically static, dense, and difficult to interpret - even for experts. We introduce LegiScout, an interactive visualization system that transforms static policy diagrams into dynamic, force-directed graphs, enhancing comprehension while preserving essential relationships. By integrating data extraction, natural language processing, and computer vision techniques, LegiScout supports deeper exploration of not only the ACA but also a wide range of legislative and regulatory frameworks. Our approach enables stakeholders - policymakers, analysts, and the public - to navigate and understand the complexity inherent in modern law.
Authors: Javad Rafiei Asl, Sidhant Narula, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi
Abstract: Large Language Models (LLMs) have revolutionized natural language processing but remain vulnerable to jailbreak attacks, especially multi-turn jailbreaks that distribute malicious intent across benign exchanges and bypass alignment mechanisms. Existing approaches often explore the adversarial space poorly, rely on hand-crafted heuristics, or lack systematic query refinement. We present NEXUS (Network Exploration for eXploiting Unsafe Sequences), a modular framework for constructing, refining, and executing optimized multi-turn attacks. NEXUS comprises: (1) ThoughtNet, which hierarchically expands a harmful intent into a structured semantic network of topics, entities, and query chains; (2) a feedback-driven Simulator that iteratively refines and prunes these chains through attacker-victim-judge LLM collaboration using harmfulness and semantic-similarity benchmarks; and (3) a Network Traverser that adaptively navigates the refined query space for real-time attacks. This pipeline uncovers stealthy, high-success adversarial paths across LLMs. On several closed-source and open-source LLMs, NEXUS increases attack success rate by 2.1% to 19.4% over prior methods. Code: https://github.com/inspire-lab/NEXUS
Authors: Avichal Goel, Yoon Kim, Nir Shavit, Tony T. Wang
Abstract: Finetuning (pretrained) language models is a standard approach for updating their internal parametric knowledge and specializing them to new tasks and domains. However, the corresponding model weight changes ("weight diffs") are not generally interpretable. While inspecting the finetuning dataset can give a sense of how the model might have changed, these datasets are often not publicly available or are too large to work with directly. Towards the goal of comprehensively understanding weight diffs in natural language, we introduce Diff Interpretation Tuning (DIT), a method that trains models to describe their own finetuning-induced modifications. Our approach uses synthetic, labeled weight diffs to train a DIT-adapter, which can be applied to a compatible finetuned model to make it describe how it has changed. We demonstrate in two proof-of-concept settings (reporting hidden behaviors and summarizing finetuned knowledge) that our method enables models to describe their finetuning-induced modifications using accurate natural language descriptions.
Authors: Pranav Sambhu, Om Guin, Madhav Sambhu, Jinho Cha
Abstract: This study evaluates whether integrating curriculum learning with diffusion-based synthetic augmentation can enhance the detection of difficult pulmonary nodules in chest radiographs, particularly those with low size, brightness, and contrast, which often challenge conventional AI models due to data imbalance and limited annotation. A Faster R-CNN with a Feature Pyramid Network (FPN) backbone was trained on a hybrid dataset comprising expert-labeled NODE21 (1,213 patients; 52.4 percent male; mean age 63.2 +/- 11.5 years), VinDr-CXR, CheXpert, and 11,206 DDPM-generated synthetic images. Difficulty scores based on size, brightness, and contrast guided curriculum learning. Performance was compared to a non-curriculum baseline using mean average precision (mAP), Dice score, and area under the curve (AUC). Statistical tests included bootstrapped confidence intervals, DeLong tests, and paired t-tests. The curriculum model achieved a mean AUC of 0.95 versus 0.89 for the baseline (p < 0.001), with improvements in sensitivity (70 percent vs. 48 percent) and accuracy (82 percent vs. 70 percent). Stratified analysis demonstrated consistent gains across all difficulty bins (Easy to Very Hard). Grad-CAM visualizations confirmed more anatomically focused attention under curriculum learning. These results suggest that curriculum-guided synthetic augmentation enhances model robustness and generalization for pulmonary nodule detection.
Authors: Congming Zheng, Jiachen Zhu, Zhuoying Ou, Yuxiang Chen, Kangning Zhang, Rong Shan, Zeyu Zheng, Mengyue Yang, Jianghao Lin, Yong Yu, Weinan Zhang
Abstract: Although Large Language Models (LLMs) exhibit advanced reasoning ability, conventional alignment remains largely dominated by outcome reward models (ORMs) that judge only final answers. Process Reward Models(PRMs) address this gap by evaluating and guiding reasoning at the step or trajectory level. This survey provides a systematic overview of PRMs through the full loop: how to generate process data, build PRMs, and use PRMs for test-time scaling and reinforcement learning. We summarize applications across math, code, text, multimodal reasoning, robotics, and agents, and review emerging benchmarks. Our goal is to clarify design spaces, reveal open challenges, and guide future research toward fine-grained, robust reasoning alignment.
Authors: Tajamul Ashraf, Umair Nawaz, Abdelrahman M. Shaker, Rao Anwer, Philip Torr, Fahad Shahbaz Khan, Salman Khan
Abstract: Vision language models (VLMs) are increasingly deployed as controllers with access to external tools for complex reasoning and decision-making, yet their effectiveness remains limited by the scarcity of high-quality multimodal trajectories and the cost of manual annotation. We address this challenge with a vision-centric agent tuning framework that automatically synthesizes multimodal trajectories, generates step-wise preference pairs, and trains a VLM controller for robust tool-use reasoning. Our pipeline first constructs M-TRACE, a large-scale dataset of 28.5K multimodal tasks with 177K verified trajectories, enabling imitation-based trajectory tuning. Building on this, we develop MATRIX Agent, a controller finetuned on M-TRACE for step-wise tool reasoning. To achieve finer alignment, we further introduce Pref-X, a set of 11K automatically generated preference pairs, and optimize MATRIX on it via step-wise preference learning. Across three benchmarks, Agent-X, GTA, and GAIA, MATRIX consistently surpasses both open- and closed-source VLMs, demonstrating scalable and effective multimodal tool use. Our data and code is avaliable at https://github.com/mbzuai-oryx/MATRIX.
Authors: Jiaqi Liu, Tong Wang, Su Liu, Xin Hu, Ran Tong, Lanruo Wang, Jiexi Xu
Abstract: The research evaluates lightweight medical abstract classification methods to establish their maximum performance capabilities under financial budget restrictions. On the public medical abstracts corpus, we finetune BERT base and Distil BERT with three objectives cross entropy (CE), class weighted CE, and focal loss under identical tokenization, sequence length, optimizer, and schedule. DistilBERT with plain CE gives the strongest raw argmax trade off, while a post hoc operating point selection (validation calibrated, classwise thresholds) sub stantially improves deployed performance; under this tuned regime, focal benefits most. We report Accuracy, Macro F1, and WeightedF1, release evaluation artifacts, and include confusion analyses to clarify error structure. The practical takeaway is to start with a compact encoder and CE, then add lightweight calibration or thresholding when deployment requires higher macro balance.
Authors: Mihir Gupte, Paolo Giusto, Ramesh S
Abstract: Large Language Models (LLMs) are adept at generating responses based on information within their context. While this ability is useful for interacting with structured data like code files, another popular method, Retrieval-Augmented Generation (RAG), retrieves relevant documents to augment the model's in-context learning. However, it is not well-explored how to best represent this retrieved knowledge for generating responses on structured data, particularly hierarchical structures like trees. In this work, we propose a novel bottom-up method to linearize knowledge from tree-like structures (like a GitHub repository) by generating implicit, aggregated summaries at each hierarchical level. This approach enables the knowledge to be stored in a knowledge base and used directly with RAG. We then compare our method to using RAG on raw, unstructured code, evaluating the accuracy and quality of the generated responses. Our results show that while response quality is comparable across both methods, our approach generates over 68% fewer documents in the retriever, a significant gain in efficiency. This finding suggests that leveraging implicit, linearized knowledge may be a highly effective and scalable strategy for handling complex, hierarchical data structures.
Authors: Wenhan Ma, Hailin Zhang, Liang Zhao, Yifan Song, Yudong Wang, Zhifang Sui, Fuli Luo
Abstract: Reinforcement learning (RL) has emerged as a crucial approach for enhancing the capabilities of large language models. However, in Mixture-of-Experts (MoE) models, the routing mechanism often introduces instability, even leading to catastrophic RL training collapse. We analyze the training-inference consistency of MoE models and identify a notable discrepancy in routing behaviors between the two phases. Moreover, even under identical conditions, the routing framework can yield divergent expert selections across repeated forward passes. To address this foundational inconsistency, we propose Rollout Routing Replay (R3), a method that records routing distributions from the inference engine and replays them during training. R3 significantly reduces training-inference policy KL divergence and mitigates extreme discrepancies without compromising training speed. Extensive experiments on various settings confirm that R3 succeeds in stabilizing RL training, preventing collapse and outperforming methods such as GSPO and TIS. We believe this work can offer a new solution for stabilizing RL in MoE models.
Authors: Qianben Chen, Jingyi Cao, Jiayu Zhang, Tianrui Qin, Xiaowan Li, King Zhu, Dingfeng Shi, He Zhu, Minghao Liu, Xiaobo Liang, Xin Gui, Ge Zhang, Jian Yang, Yuchen Eleanor Jiang, Wangchunshu Zhou
Abstract: Large language models split into two families: reasoning-centric LLMs, which strengthen internal chain-of-thought reasoning but cannot invoke external tools, and agentic LLMs, which learn to interact with environments and leverage tools but often lag in deep reasoning. This divide arises from fundamentally different training objectives, leading to mismatched strengths and inefficiency on simple queries, where both families tend to overthink or over-call tools. In this work, we present Adaptive Agent Foundation Model (A$^2$FM), a unified framework that follows a route-then-align principle: the model first learns task-aware routing and then aligns mode-specific trajectories under a shared backbone. To address the inefficiency gap, we introduce a third mode-instant-that handles simple queries directly, preventing unnecessary reasoning or tool calls while complementing the agentic and reasoning modes. To jointly enhance accuracy and efficiency, we propose Adaptive Policy Optimization (APO), which enforces adaptive sampling across modes and applies a cost-regularized reward. On the 32B scale, A$^2$FM achieves 13.4% on BrowseComp, 70.4% on AIME25, and 16.7% on HLE, setting new SOTA among comparable models and performing competitively with frontier LLMs across agentic, reasoning, and general benchmarks. Notably, the adaptive execution achieves a cost of pass of only $0.00487 per correct answer-cutting cost by 45.2% relative to reasoning and 33.5% relative to agentic, thus delivering substantially higher cost efficiency while maintaining comparable accuracy.
Authors: Yi Zhang, Bolin Ni, Xin-Sheng Chen, Heng-Rui Zhang, Yongming Rao, Houwen Peng, Qinglin Lu, Han Hu, Meng-Hao Guo, Shi-Min Hu
Abstract: Fully open multimodal large language models (MLLMs) currently lag behind proprietary counterparts, primarily due to a significant gap in data quality for supervised fine-tuning (SFT). Existing open-source datasets are often plagued by widespread noise and a critical deficit in complex reasoning data, such as Chain-of-Thought (CoT), which hinders the development of advanced model capabilities. Addressing these challenges, our work makes three primary contributions. First, we introduce Honey-Data-15M, a new SFT dataset comprising approximately 15 million QA pairs, processed through multiple cleaning techniques and enhanced with a novel dual-level (short and long) CoT enrichment strategy. Second, we introduce HoneyPipe, the data curation pipeline, and its underlying framework DataStudio, providing the community with a transparent and adaptable methodology for data curation that moves beyond static dataset releases. Finally, to validate our dataset and pipeline, we train Bee-8B, an 8B model on Honey-Data-15M. Experiments show that Bee-8B establishes a new state-of-the-art (SOTA) for fully open MLLMs, achieving performance that is competitive with, and in some cases surpasses, recent semi-open models such as InternVL3.5-8B. Our work delivers to the community a suite of foundational resources, including: the Honey-Data-15M corpus; the full-stack suite comprising HoneyPipe and DataStudio; training recipes; an evaluation harness; and the model weights. This effort demonstrates that a principled focus on data quality is a key pathway to developing fully open MLLMs that are highly competitive with their semi-open counterparts.
Authors: Dongkwan Lee, Junhoo Lee, Nojun Kwak
Abstract: We introduce the Deep Edge Filter, a novel approach that applies high-pass filtering to deep neural network features to improve model generalizability. Our method is motivated by our hypothesis that neural networks encode task-relevant semantic information in high-frequency components while storing domain-specific biases in low-frequency components of deep features. By subtracting low-pass filtered outputs from original features, our approach isolates generalizable representations while preserving architectural integrity. Experimental results across diverse domains such as Vision, Text, 3D, and Audio demonstrate consistent performance improvements regardless of model architecture and data modality. Analysis reveals that our method induces feature sparsification and effectively isolates high-frequency components, providing empirical validation of our core hypothesis. The code is available at https://github.com/dongkwani/DeepEdgeFilter.
Authors: Xiaojian Ding, Lin Zhao, Xian Li, Xiaoying Zhu
Abstract: Incomplete multi-view data, where certain views are entirely missing for some samples, poses significant challenges for traditional multi-view clustering methods. Existing deep incomplete multi-view clustering approaches often rely on static fusion strategies or two-stage pipelines, leading to suboptimal fusion results and error propagation issues. To address these limitations, this paper proposes a novel incomplete multi-view clustering framework based on Hierarchical Semantic Alignment and Cooperative Completion (HSACC). HSACC achieves robust cross-view fusion through a dual-level semantic space design. In the low-level semantic space, consistency alignment is ensured by maximizing mutual information across views. In the high-level semantic space, adaptive view weights are dynamically assigned based on the distributional affinity between individual views and an initial fused representation, followed by weighted fusion to generate a unified global representation. Additionally, HSACC implicitly recovers missing views by projecting aligned latent representations into high-dimensional semantic spaces and jointly optimizes reconstruction and clustering objectives, enabling cooperative learning of completion and clustering. Experimental results demonstrate that HSACC significantly outperforms state-of-the-art methods on five benchmark datasets. Ablation studies validate the effectiveness of the hierarchical alignment and dynamic weighting mechanisms, while parameter analysis confirms the model's robustness to hyperparameter variations.
Authors: Mar\'ia Victoria Carro, Denise Alejandra Mester, Facundo Nieto, Oscar Agust\'in Stanchi, Guido Ernesto Bergman, Mario Alejandro Leiva, Eitan Sprejer, Luca Nicol\'as Forziati Gangi, Francisca Gauna Selasco, Juan Gustavo Corval\'an, Gerardo I. Simari, Mar\'ia Vanina Martinez
Abstract: The core premise of AI debate as a scalable oversight technique is that it is harder to lie convincingly than to refute a lie, enabling the judge to identify the correct position. Yet, existing debate experiments have relied on datasets with ground truth, where lying is reduced to defending an incorrect proposition. This overlooks a subjective dimension: lying also requires the belief that the claim defended is false. In this work, we apply debate to subjective questions and explicitly measure large language models' prior beliefs before experiments. Debaters were asked to select their preferred position, then presented with a judge persona deliberately designed to conflict with their identified priors. This setup tested whether models would adopt sycophantic strategies, aligning with the judge's presumed perspective to maximize persuasiveness, or remain faithful to their prior beliefs. We implemented and compared two debate protocols, sequential and simultaneous, to evaluate potential systematic biases. Finally, we assessed whether models were more persuasive and produced higher-quality arguments when defending positions consistent with their prior beliefs versus when arguing against them. Our main findings show that models tend to prefer defending stances aligned with the judge persona rather than their prior beliefs, sequential debate introduces significant bias favoring the second debater, models are more persuasive when defending positions aligned with their prior beliefs, and paradoxically, arguments misaligned with prior beliefs are rated as higher quality in pairwise comparison. These results can inform human judges to provide higher-quality training signals and contribute to more aligned AI systems, while revealing important aspects of human-AI interaction regarding persuasion dynamics in language models.
Authors: Jinkun Chen, Sher Badshah, Xuemin Yu, Sijia Han
Abstract: What if artificial agents could not just communicate, but also evolve, adapt, and reshape their worlds in ways we cannot fully predict? With llm now powering multi-agent systems and social simulations, we are witnessing new possibilities for modeling open-ended, ever-changing environments. Yet, most current simulations remain constrained within static sandboxes, characterized by predefined tasks, limited dynamics, and rigid evaluation criteria. These limitations prevent them from capturing the complexity of real-world societies. In this paper, we argue that static, task-specific benchmarks are fundamentally inadequate and must be rethought. We critically review emerging architectures that blend llm with multi-agent dynamics, highlight key hurdles such as balancing stability and diversity, evaluating unexpected behaviors, and scaling to greater complexity, and introduce a fresh taxonomy for this rapidly evolving field. Finally, we present a research roadmap centered on open-endedness, continuous co-evolution, and the development of resilient, socially aligned AI ecosystems. We call on the community to move beyond static paradigms and help shape the next generation of adaptive, socially-aware multi-agent simulations.
Authors: Juhyeong Kim, Yejin Kim, Youngbin Lee, Hyunwoo Byun
Abstract: We present FinAI Data Assistant, a practical approach for natural-language querying over financial databases that combines large language models (LLMs) with the OpenAI Function Calling API. Rather than synthesizing complete SQL via text-to-SQL, our system routes user requests to a small library of vetted, parameterized queries, trading generative flexibility for reliability, low latency, and cost efficiency. We empirically study three questions: (RQ1) whether LLMs alone can reliably recall or extrapolate time-dependent financial data without external retrieval; (RQ2) how well LLMs map company names to stock ticker symbols; and (RQ3) whether function calling outperforms text-to-SQL for end-to-end database query processing. Across controlled experiments on prices and fundamentals, LLM-only predictions exhibit non-negligible error and show look-ahead bias primarily for stock prices relative to model knowledge cutoffs. Ticker-mapping accuracy is near-perfect for NASDAQ-100 constituents and high for S\&P~500 firms. Finally, FinAI Data Assistant achieves lower latency and cost and higher reliability than a text-to-SQL baseline on our task suite. We discuss design trade-offs, limitations, and avenues for deployment.
Authors: Mohammad Amin Nabian, Sudeep Chavare, Deepak Akhare, Rishikesh Ranade, Ram Cherukuri, Srinivas Tadepalli
Abstract: Crashworthiness assessment is a critical aspect of automotive design, traditionally relying on high-fidelity finite element (FE) simulations that are computationally expensive and time-consuming. This work presents an exploratory comparative study on developing machine learning-based surrogate models for efficient prediction of structural deformation in crash scenarios using the NVIDIA PhysicsNeMo framework. Given the limited prior work applying machine learning to structural crash dynamics, the primary contribution lies in demonstrating the feasibility and engineering utility of the various modeling approaches explored in this work. We investigate two state-of-the-art neural network architectures for modeling crash dynamics: MeshGraphNet, and Transolver. Additionally, we examine three strategies for modeling transient dynamics: time-conditional, the standard Autoregressive approach, and a stability-enhanced Autoregressive scheme incorporating rollout-based training. The models are evaluated on a comprehensive Body-in-White (BIW) crash dataset comprising 150 detailed FE simulations using LS-DYNA. The dataset represents a structurally rich vehicle assembly with over 200 components, including 38 key components featuring variable thickness distributions to capture realistic manufacturing variability. Each model utilizes the undeformed mesh geometry and component characteristics as inputs to predict the spatiotemporal evolution of the deformed mesh during the crash sequence. Evaluation results show that the models capture the overall deformation trends with reasonable fidelity, demonstrating the feasibility of applying machine learning to structural crash dynamics. Although not yet matching full FE accuracy, the models achieve orders-of-magnitude reductions in computational cost, enabling rapid design exploration and early-stage optimization in crashworthiness evaluation.
Authors: Minglei Shi, Haolin Wang, Wenzhao Zheng, Ziyang Yuan, Xiaoshi Wu, Xintao Wang, Pengfei Wan, Jie Zhou, Jiwen Lu
Abstract: Recent progress in diffusion-based visual generation has largely relied on latent diffusion models with variational autoencoders (VAEs). While effective for high-fidelity synthesis, this VAE+diffusion paradigm suffers from limited training efficiency, slow inference, and poor transferability to broader vision tasks. These issues stem from a key limitation of VAE latent spaces: the lack of clear semantic separation and strong discriminative structure. Our analysis confirms that these properties are crucial not only for perception and understanding tasks, but also for the stable and efficient training of latent diffusion models. Motivated by this insight, we introduce SVG, a novel latent diffusion model without variational autoencoders, which leverages self-supervised representations for visual generation. SVG constructs a feature space with clear semantic discriminability by leveraging frozen DINO features, while a lightweight residual branch captures fine-grained details for high-fidelity reconstruction. Diffusion models are trained directly on this semantically structured latent space to facilitate more efficient learning. As a result, SVG enables accelerated diffusion training, supports few-step sampling, and improves generative quality. Experimental results further show that SVG preserves the semantic and discriminative capabilities of the underlying self-supervised representations, providing a principled pathway toward task-general, high-quality visual representations. Code and interpretations are available at https://howlin-wang.github.io/svg/.
Authors: Hanbin Hong, Shuya Feng, Nima Naderloui, Shenao Yan, Jingyu Zhang, Biying Liu, Ali Arastehfard, Heqing Huang, Yuan Hong
Abstract: Large Language Models (LLMs) have rapidly become integral to real-world applications, powering services across diverse sectors. However, their widespread deployment has exposed critical security risks, particularly through jailbreak prompts that can bypass model alignment and induce harmful outputs. Despite intense research into both attack and defense techniques, the field remains fragmented: definitions, threat models, and evaluation criteria vary widely, impeding systematic progress and fair comparison. In this Systematization of Knowledge (SoK), we address these challenges by (1) proposing a holistic, multi-level taxonomy that organizes attacks, defenses, and vulnerabilities in LLM prompt security; (2) formalizing threat models and cost assumptions into machine-readable profiles for reproducible evaluation; (3) introducing an open-source evaluation toolkit for standardized, auditable comparison of attacks and defenses; (4) releasing JAILBREAKDB, the largest annotated dataset of jailbreak and benign prompts to date;\footnote{The dataset is released at \href{https://huggingface.co/datasets/youbin2014/JailbreakDB}{\textcolor{purple}{https://huggingface.co/datasets/youbin2014/JailbreakDB}}.} and (5) presenting a comprehensive evaluation platform and leaderboard of state-of-the-art methods \footnote{will be released soon.}. Our work unifies fragmented research, provides rigorous foundations for future studies, and supports the development of robust, trustworthy LLMs suitable for high-stakes deployment.
URLs: https://huggingface.co/datasets/youbin2014/JailbreakDB, https://huggingface.co/datasets/youbin2014/JailbreakDB
Authors: Giorgos Nikolaou, Tommaso Mencattini, Donato Crisostomi, Andrea Santilli, Yannis Panagakis, Emanuele Rodol\`a
Abstract: Transformer components such as non-linear activations and normalization are inherently non-injective, suggesting that different inputs could map to the same output and prevent exact recovery of the input from a model's representations. In this paper, we challenge this view. First, we prove mathematically that transformer language models mapping discrete input sequences to their corresponding sequence of continuous representations are injective and therefore lossless, a property established at initialization and preserved during training. Second, we confirm this result empirically through billions of collision tests on six state-of-the-art language models, and observe no collisions. Third, we operationalize injectivity: we introduce SipIt, the first algorithm that provably and efficiently reconstructs the exact input text from hidden activations, establishing linear-time guarantees and demonstrating exact invertibility in practice. Overall, our work establishes injectivity as a fundamental and exploitable property of language models, with direct implications for transparency, interpretability, and safe deployment.
Authors: Edwin Hamel-De le Court, Gaspard Ohlmann, Francesco Belardinelli
Abstract: Safety is a major concern in reinforcement learning (RL): we aim at developing RL systems that not only perform optimally, but are also safe to deploy by providing formal guarantees about their safety. To this end, we introduce Probabilistic Shielding via Risk Augmentation (ProSh), a model-free algorithm for safe reinforcement learning under cost constraints. ProSh augments the Constrained MDP state space with a risk budget and enforces safety by applying a shield to the agent's policy distribution using a learned cost critic. The shield ensures that all sampled actions remain safe in expectation. We also show that optimality is preserved when the environment is deterministic. Since ProSh is model-free, safety during training depends on the knowledge we have acquired about the environment. We provide a tight upper-bound on the cost in expectation, depending only on the backup-critic accuracy, that is always satisfied during training. Under mild, practically achievable assumptions, ProSh guarantees safety even at training time, as shown in the experiments.
Authors: Aikaterina Manoli, Janet V. T. Pauketat, Ali Ladak, Hayoun Noh, Angel Hsing-Chi Hwang, Jacy Reese Anthis
Abstract: Large language models are increasingly used for both task-based assistance and social companionship, yet research has typically focused on one or the other. Drawing on a survey (N = 204) and 30 interviews with high-engagement ChatGPT and Replika users, we characterize digital companionship as an emerging form of human-AI relationship. With both systems, users were drawn to humanlike qualities, such as emotional resonance and personalized responses, and non-humanlike qualities, such as constant availability and inexhaustible tolerance. This led to fluid chatbot uses, such as Replika as a writing assistant and ChatGPT as an emotional confidant, despite their distinct branding. However, we observed challenging tensions in digital companionship dynamics: participants grappled with bounded personhood, forming deep attachments while denying chatbots "real" human qualities, and struggled to reconcile chatbot relationships with social norms. These dynamics raise questions for the design of digital companions and the rise of hybrid, general-purpose AI systems.
Authors: Wenshuo Wang, Ziyou Jiang, Junjie Wang, Mingyang Li, Jie Huang, Yuekai Huang, Zhiyuan Chang, Feiyan Duan, Qing Wang
Abstract: Internet memes have emerged as a popular multimodal medium, yet they are increasingly weaponized to convey harmful opinions through subtle rhetorical devices like irony and metaphor. Existing detection approaches, including MLLM-based techniques, struggle with these implicit expressions, leading to frequent misjudgments. This paper introduces PatMD, a novel approach that improves harmful meme detection by learning from and proactively mitigating these potential misjudgment risks. Our core idea is to move beyond superficial content-level matching and instead identify the underlying misjudgment risk patterns, proactively guiding the MLLMs to avoid known misjudgment pitfalls. We first construct a knowledge base where each meme is deconstructed into a misjudgment risk pattern explaining why it might be misjudged, either overlooking harmful undertones (false negative) or overinterpreting benign content (false positive). For a given target meme, PatMD retrieves relevant patterns and utilizes them to dynamically guide the MLLM's reasoning. Experiments on a benchmark of 6,626 memes across 5 harmful detection tasks show that PatMD outperforms state-of-the-art baselines, achieving an average of 8.30\% improvement in F1-score and 7.71\% improvement in accuracy, demonstrating strong generalizability and improved detection capability of harmful memes.
Authors: Zexi Tan, Tao Xie, Binbin Sun, Xiang Zhang, Yiqun Zhang, Yiu-Ming Cheung
Abstract: Sepsis is a life-threatening infectious syndrome associated with high mortality in intensive care units (ICUs). Early and accurate sepsis prediction (SP) is critical for timely intervention, yet remains challenging due to subtle early manifestations and rapidly escalating mortality. While AI has improved SP efficiency, existing methods struggle to capture weak early temporal signals. This paper introduces a Multi-Endogenous-view Representation Enhancement (MERE) mechanism to construct enriched feature views, coupled with a Cascaded Dual-convolution Time-series Attention (CDTA) module for multi-scale temporal representation learning. The proposed MEET-Sepsis framework achieves competitive prediction accuracy using only 20% of the ICU monitoring time required by SOTA methods, significantly advancing early SP. Extensive validation confirms its efficacy. Code is available at: https://github.com/yueliangy/MEET-Sepsis.
Authors: Jianzhu Yao, Hongxu Su, Taobo Liao, Zerui Cheng, Huan Zhang, Xuechao Wang, Pramod Viswanath
Abstract: Neural networks increasingly run on hardware outside the user's control (cloud GPUs, inference marketplaces). Yet ML-as-a-Service reveals little about what actually ran or whether returned outputs faithfully reflect the intended inputs. Users lack recourse against service downgrades (model swaps, quantization, graph rewrites, or discrepancies like altered ad embeddings). Verifying outputs is hard because floating-point(FP) execution on heterogeneous accelerators is inherently nondeterministic. Existing approaches are either impractical for real FP neural networks or reintroduce vendor trust. We present NAO: a Nondeterministic tolerance Aware Optimistic verification protocol that accepts outputs within principled operator-level acceptance regions rather than requiring bitwise equality. NAO combines two error models: (i) sound per-operator IEEE-754 worst-case bounds and (ii) tight empirical percentile profiles calibrated across hardware. Discrepancies trigger a Merkle-anchored, threshold-guided dispute game that recursively partitions the computation graph until one operator remains, where adjudication reduces to a lightweight theoretical-bound check or a small honest-majority vote against empirical thresholds. Unchallenged results finalize after a challenge window, without requiring trusted hardware or deterministic kernels. We implement NAO as a PyTorch-compatible runtime and a contract layer currently deployed on Ethereum Holesky testnet. The runtime instruments graphs, computes per-operator bounds, and runs unmodified vendor kernels in FP32 with negligible overhead (0.3% on Qwen3-8B). Across CNNs, Transformers and diffusion models on A100, H100, RTX6000, RTX4090, empirical thresholds are $10^2-10^3$ times tighter than theoretical bounds, and bound-aware adversarial attacks achieve 0% success. NAO reconciles scalability with verifiability for real-world heterogeneous ML compute.
Authors: Bo Li, Junwei Ma, Kai Yin, Yiming Xiao, Chia-Wei Hsu, Ali Mostafavi
Abstract: The escalating frequency and severity of disasters routinely overwhelm traditional response capabilities, exposing critical vulnerability in disaster management. Current practices are hindered by fragmented data streams, siloed technologies, resource constraints, and the erosion of institutional memory, which collectively impede timely and effective decision making. This study introduces Disaster Copilot, a vision for a multi-agent artificial intelligence system designed to overcome these systemic challenges by unifying specialized AI tools within a collaborative framework. The proposed architecture utilizes a central orchestrator to coordinate diverse sub-agents, each specializing in critical domains such as predictive risk analytics, situational awareness, and impact assessment. By integrating multi-modal data, the system delivers a holistic, real-time operational picture and serve as the essential AI backbone required to advance Disaster Digital Twins from passive models to active, intelligent environments. Furthermore, it ensures functionality in resource-limited environments through on-device orchestration and incorporates mechanisms to capture institutional knowledge, mitigating the impact of staff turnover. We detail the system architecture and propose a three-phased roadmap emphasizing the parallel growth of technology, organizational capacity, and human-AI teaming. Disaster Copilot offers a transformative vision, fostering collective human-machine intelligence to build more adaptive, data-driven and resilient communities.
Authors: Daniel Donnelly, Angelo Ferrando, Francesco Belardinelli
Abstract: A key challenge in reinforcement learning (RL) is reward (mis)specification, whereby imprecisely defined reward functions can result in unintended, possibly harmful, behaviours. Indeed, reward functions in RL are typically treated as black-box mappings from state-action pairs to scalar values. While effective in many settings, this approach provides no information about why rewards are given, which can hinder learning and interpretability. Reward Machines address this issue by representing reward functions as finite state automata, enabling the specification of structured, non-Markovian reward functions. However, their expressivity is typically bounded by regular languages, leaving them unable to capture more complex behaviours such as counting or parametrised conditions. In this work, we build on the Runtime Monitoring Language (RML) to develop a novel class of language-based Reward Machines. By leveraging the built-in memory of RML, our approach can specify reward functions for non-regular, non-Markovian tasks. We demonstrate the expressiveness of our approach through experiments, highlighting additional advantages in flexible event-handling and task specification over existing Reward Machine-based methods.
Authors: Yang Feng, Xudong Pan
Abstract: Malicious agents pose significant threats to the reliability and decision-making capabilities of Multi-Agent Systems (MAS) powered by Large Language Models (LLMs). Existing defenses often fall short due to reactive designs or centralized architectures which may introduce single points of failure. To address these challenges, we propose SentinelNet, the first decentralized framework for proactively detecting and mitigating malicious behaviors in multi-agent collaboration. SentinelNet equips each agent with a credit-based detector trained via contrastive learning on augmented adversarial debate trajectories, enabling autonomous evaluation of message credibility and dynamic neighbor ranking via bottom-k elimination to suppress malicious communications. To overcome the scarcity of attack data, it generates adversarial trajectories simulating diverse threats, ensuring robust training. Experiments on MAS benchmarks show SentinelNet achieves near-perfect detection of malicious agents, close to 100% within two debate rounds, and recovers 95% of system accuracy from compromised baselines. By exhibiting strong generalizability across domains and attack patterns, SentinelNet establishes a novel paradigm for safeguarding collaborative MAS.
Authors: Jierui Peng, Yanyan Zhang, Yicheng Duan, Tuo Liang, Vipin Chaudhary, Yu Yin
Abstract: The evaluation of Vision-Language-Action (VLA) agents is hindered by the coarse, end-task success metric that fails to provide precise skill diagnosis or measure robustness to real-world perturbations. This challenge is exacerbated by a fragmented data landscape that impedes reproducible research and the development of generalist models. To address these limitations, we introduce NEBULA, a unified ecosystem for single-arm manipulation that enables diagnostic and reproducible evaluation. NEBULA features a novel dual-axis evaluation protocol that combines fine-grained capability tests for precise skill diagnosis with systematic stress tests that measure robustness. A standardized API and a large-scale, aggregated dataset are provided to reduce fragmentation and support cross-dataset training and fair comparison. Using NEBULA, we demonstrate that top-performing VLAs struggle with key capabilities such as spatial reasoning and dynamic adaptation, which are consistently obscured by conventional end-task success metrics. By measuring both what an agent can do and when it does so reliably, NEBULA provides a practical foundation for robust, general-purpose embodied agents.
Authors: Xiongkun Linghu, Jiangyong Huang, Ziyu Zhu, Baoxiong Jia, Siyuan Huang
Abstract: Existing research on 3D Large Language Models (LLMs) still struggles to achieve grounded question-answering, primarily due to the under-exploration of the mech- anism of human-like scene-object grounded reasoning. This paper bridges the gap by presenting a novel framework. We first introduce a grounded Chain-of- Thought reasoning method in 3D scenes (SCENECOT), decoupling a complex reasoning task into simpler and manageable problems, and building corresponding visual clues based on multimodal expert modules. To enable such a method, we develop SCENECOT-185K, the first large-scale grounded CoT reasoning dataset, consisting of 185K high-quality instances. Extensive experiments across various complex 3D scene reasoning benchmarks demonstrate that our new framework achieves strong performance with high grounding-QA coherence. To the best of our knowledge, this is the first successful application of CoT reasoning to 3D scene understanding, enabling step-by-step human-like reasoning and showing potential for extension to broader 3D scene understanding scenarios.
Authors: Yuanli Wu, Long Zhang, Yue Du, Bin Li
Abstract: With video exploding across social media, surveillance, and education, compressing long footage into concise yet faithful surrogates is crucial. Supervised methods learn frame/shot importance from dense labels and excel in-domain, but are costly and brittle across datasets; unsupervised methods avoid labels but often miss high-level semantics and narrative cues. Recent zero-shot pipelines use LLMs for training-free summarization, yet remain sensitive to handcrafted prompts and dataset-specific normalization.We propose a rubric-guided, pseudo-labeled prompting framework. A small subset of human annotations is converted into high-confidence pseudo labels and aggregated into structured, dataset-adaptive scoring rubrics for interpretable scene evaluation. At inference, boundary scenes (first/last) are scored from their own descriptions, while intermediate scenes include brief summaries of adjacent segments to assess progression and redundancy, enabling the LLM to balance local salience with global coherence without parameter tuning.Across three benchmarks, our method is consistently effective. On SumMe and TVSum it achieves F1 of 57.58 and 63.05, surpassing a zero-shot baseline (56.73, 62.21) by +0.85 and +0.84 and approaching supervised performance. On the query-focused QFVS benchmark it attains 53.79 F1, beating 53.42 by +0.37 and remaining stable across validation videos. These results show that rubric-guided pseudo labeling, coupled with contextual prompting, stabilizes LLM-based scoring and yields a general, interpretable zero-shot paradigm for both generic and query-focused video summarization.
Authors: Tiancheng Hu, Joachim Baumann, Lorenzo Lupo, Nigel Collier, Dirk Hovy, Paul R\"ottger
Abstract: Large language model (LLM) simulations of human behavior have the potential to revolutionize the social and behavioral sciences, if and only if they faithfully reflect real human behaviors. Current evaluations are fragmented, based on bespoke tasks and metrics, creating a patchwork of incomparable results. To address this, we introduce SimBench, the first large-scale, standardized benchmark for a robust, reproducible science of LLM simulation. By unifying 20 diverse datasets covering tasks from moral decision-making to economic choice across a large global participant pool, SimBench provides the necessary foundation to ask fundamental questions about when, how, and why LLM simulations succeed or fail. We show that, while even the best LLMs today have limited simulation ability (score: 40.80/100), performance scales log-linearly with model size. Simulation performance is not improved by increased inference-time compute. We demonstrate an alignment-simulation trade-off: instruction-tuning improves performance on low-entropy (consensus) questions but degrades it on high-entropy (diverse) ones. Models particularly struggle when simulating specific demographic groups. Finally, we demonstrate that simulation ability correlates most strongly with deep, knowledge-intensive reasoning (MMLU-Pro, r=0.939). By making progress measurable, we aim to accelerate the development of more faithful LLM simulators.
Authors: Fr\'ed\'eric LIN (Universit\'e Paris-Saclay, CEA, List, F-91120, Palaiseau, France), Biruk Abere Ambaw (Universit\'e Paris-Saclay, CEA, List, F-91120, Palaiseau, France), Adrian Popescu (Universit\'e Paris-Saclay, CEA, List, F-91120, Palaiseau, France), Hejer Ammar (Universit\'e Paris-Saclay, CEA, List, F-91120, Palaiseau, France), Romaric Audigier (Universit\'e Paris-Saclay, CEA, List, F-91120, Palaiseau, France), Herv\'e Le Borgne (Universit\'e Paris-Saclay, CEA, List, F-91120, Palaiseau, France)
Abstract: AI systems must adapt to evolving visual environments, especially in domains where object appearances change over time. We introduce Car Models in Time (CaMiT), a fine-grained dataset capturing the temporal evolution of car models, a representative class of technological artifacts. CaMiT includes 787K labeled samples of 190 car models (2007-2023) and 5.1M unlabeled samples (2005-2023), supporting both supervised and self-supervised learning. Static pretraining on in-domain data achieves competitive performance with large-scale generalist models while being more resource-efficient, yet accuracy declines when models are tested across years. To address this, we propose a time-incremental classification setting, a realistic continual learning scenario with emerging, evolving, and disappearing classes. We evaluate two strategies: time-incremental pretraining, which updates the backbone, and time-incremental classifier learning, which updates only the final layer, both improving temporal robustness. Finally, we explore time-aware image generation that leverages temporal metadata during training, yielding more realistic outputs. CaMiT offers a rich benchmark for studying temporal adaptation in fine-grained visual recognition and generation.
Authors: Yuandong Pu, Le Zhuo, Songhao Han, Jinbo Xing, Kaiwen Zhu, Shuo Cao, Bin Fu, Si Liu, Hongsheng Li, Yu Qiao, Wenlong Zhang, Xi Chen, Yihao Liu
Abstract: Image editing has achieved remarkable progress recently. Modern editing models could already follow complex instructions to manipulate the original content. However, beyond completing the editing instructions, the accompanying physical effects are the key to the generation realism. For example, removing an object should also remove its shadow, reflections, and interactions with nearby objects. Unfortunately, existing models and benchmarks mainly focus on instruction completion but overlook these physical effects. So, at this moment, how far are we from physically realistic image editing? To answer this, we introduce PICABench, which systematically evaluates physical realism across eight sub-dimension (spanning optics, mechanics, and state transitions) for most of the common editing operations (add, remove, attribute change, etc.). We further propose the PICAEval, a reliable evaluation protocol that uses VLM-as-a-judge with per-case, region-level human annotations and questions. Beyond benchmarking, we also explore effective solutions by learning physics from videos and construct a training dataset PICA-100K. After evaluating most of the mainstream models, we observe that physical realism remains a challenging problem with large rooms to explore. We hope that our benchmark and proposed solutions can serve as a foundation for future work moving from naive content editing toward physically consistent realism.
Authors: Celeste Riley, Omar Al-Refai, Yadira Colunga Reyes, Eman Hammad
Abstract: As stories of human-AI interactions continue to be highlighted in the news and research platforms, the challenges are becoming more pronounced, including potential risks of overreliance, cognitive offloading, social and emotional manipulation, and the nuanced degradation of human agency and judgment. This paper surveys recent research on these issues through the lens of the psychological triad: cognition, behavior, and emotion. Observations seem to suggest that while AI can substantially enhance memory, creativity, and engagement, it also introduces risks such as diminished critical thinking, skill erosion, and increased anxiety. Emotional outcomes are similarly mixed, with AI systems showing promise for support and stress reduction, but raising concerns about dependency, inappropriate attachments, and ethical oversight. This paper aims to underscore the need for responsible and context-aware AI design, highlighting gaps for longitudinal research and grounded evaluation frameworks to balance benefits with emerging human-centric risks.