Authors: John Beverley, Regina Hurley
Abstract: This work lays the foundations for a rigorous ontological characterization of love, addressing its philosophical complexity and scientific relevance, with particular emphasis on psychology and sociology, as well as highlighting ways in which such characterization enhances relevant AI based applications. The position defended here is that love is best understood as a concatenation of passive sensations (e.g., emotional arousal) and active evaluative judgments (e.g., perceiving the beloved as valuable), in the interest of balancing the involuntary aspects of love with its rational accountability. To provide a structured foundation, the paper draws on Basic Formal Ontology (BFO) and other applied ontological methods to differentiate various senses of love. This work engages with objections to the understanding of love as concatenation, particularly concerning the relationship between sensation and judgment. A causal correlation model is defended, ensuring that the affective and cognitive components are linked. By offering a precise and scalable ontological account, this work lays the foundation for future interdisciplinary applications, making love a subject of formal inquiry in ontology engineering, artificial intelligence, and the sciences.
Authors: Kristy Wedel
Abstract: A critical challenge remains unresolved as generative AI systems are quickly implemented in various organizational settings. Despite significant advances in memory components such as RAG, vector stores, and LLM agents, these systems still have substantial memory limitations. Gen AI workflows rarely store or reflect on the full context in which decisions are made. This leads to repeated errors and a general lack of clarity. This paper introduces Contextual Memory Intelligence (CMI) as a new foundational paradigm for building intelligent systems. It repositions memory as an adaptive infrastructure necessary for longitudinal coherence, explainability, and responsible decision-making rather than passive data. Drawing on cognitive science, organizational theory, human-computer interaction, and AI governance, CMI formalizes the structured capture, inference, and regeneration of context as a fundamental system capability. The Insight Layer is presented in this paper to operationalize this vision. This modular architecture uses human-in-the-loop reflection, drift detection, and rationale preservation to incorporate contextual memory into systems. The paper argues that CMI allows systems to reason with data, history, judgment, and changing context, thereby addressing a foundational blind spot in current AI architectures and governance efforts. A framework for creating intelligent systems that are effective, reflective, auditable, and socially responsible is presented through CMI. This enhances human-AI collaboration, generative AI design, and the resilience of the institutions.
Authors: Andrei T. Patrascu
Abstract: We introduce a novel learning and planning framework that replaces traditional reward-based optimisation with constructive logical inference. In our model, actions, transitions, and goals are represented as logical propositions, and decision-making proceeds by building constructive proofs under intuitionistic logic. This method ensures that state transitions and policies are accepted only when supported by verifiable preconditions -- eschewing probabilistic trial-and-error in favour of guaranteed logical validity. We implement a symbolic agent operating in a structured gridworld, where reaching a goal requires satisfying a chain of intermediate subgoals (e.g., collecting keys to open doors), each governed by logical constraints. Unlike conventional reinforcement learning agents, which require extensive exploration and suffer from unsafe or invalid transitions, our constructive agent builds a provably correct plan through goal chaining, condition tracking, and knowledge accumulation. Empirical comparison with Q-learning demonstrates that our method achieves perfect safety, interpretable behaviour, and efficient convergence with no invalid actions, highlighting its potential for safe planning, symbolic cognition, and trustworthy AI. This work presents a new direction for reinforcement learning grounded not in numeric optimisation, but in constructive logic and proof theory.
Authors: Cecil Pang
Abstract: Contemporary businesses operate in dynamic environments requiring rapid adaptation to achieve goals and maintain competitiveness. Existing data platforms often fall short by emphasizing tools over alignment with business needs, resulting in inefficiencies and delays. To address this gap, I propose the Business Semantics Centric, AI Agents Assisted Data System (BSDS), a holistic system that integrates architecture, workflows, and team organization to ensure data systems are tailored to business priorities rather than dictated by technical constraints. BSDS redefines data systems as dynamic enablers of business success, transforming them from passive tools into active drivers of organizational growth. BSDS has a modular architecture that comprises curated data linked to business entities, a knowledge base for context-aware AI agents, and efficient data pipelines. AI agents play a pivotal role in assisting with data access and system management, reducing human effort, and improving scalability. Complementing this architecture, BSDS incorporates workflows optimized for both exploratory data analysis and production requirements, balancing speed of delivery with quality assurance. A key innovation of BSDS is its incorporation of the human factor. By aligning data team expertise with business semantics, BSDS bridges the gap between technical capabilities and business needs. Validated through real-world implementation, BSDS accelerates time-to-market for data-driven initiatives, enhances cross-functional collaboration, and provides a scalable blueprint for businesses of all sizes. Future research can build on BSDS to explore optimization strategies using complex systems and adaptive network theories, as well as developing autonomous data systems leveraging AI agents.
Authors: Rodney Sanchez, Ferat Sahin, Alexander Ororbia, Jamison Heard
Abstract: Biological and psychological concepts have inspired reinforcement learning algorithms to create new complex behaviors that expand agents' capacity. These behaviors can be seen in the rise of techniques like goal decomposition, curriculum, and intrinsic rewards, which have paved the way for these complex behaviors. One limitation in evaluating these methods is the requirement for engineered extrinsic for realistic environments. A central challenge in engineering the necessary reward function(s) comes from these environments containing states that carry high negative rewards, but provide no feedback to the agent. Death is one such stimuli that fails to provide direct feedback to the agent. In this work, we introduce an intrinsic reward function inspired by early amygdala development and produce this intrinsic reward through a novel memory-augmented neural network (MANN) architecture. We show how this intrinsic motivation serves to deter exploration of terminal states and results in avoidance behavior similar to fear conditioning observed in animals. Furthermore, we demonstrate how modifying a threshold where the fear response is active produces a range of behaviors that are described under the paradigm of general anxiety disorders (GADs). We demonstrate this behavior in the Miniworld Sidewalk environment, which provides a partially observable Markov decision process (POMDP) and a sparse reward with a non-descriptive terminal condition, i.e., death. In effect, this study results in a biologically-inspired neural architecture and framework for fear conditioning paradigms; we empirically demonstrate avoidance behavior in a constructed agent that is able to solve environments with non-descriptive terminal conditions.
Authors: Quan Shi, Carlos E. Jimenez, Shunyu Yao, Nick Haber, Diyi Yang, Karthik Narasimhan
Abstract: Recent advancements in AI reasoning have driven substantial improvements across diverse tasks. A critical open question is whether these improvements also yields better knowledge transfer: the ability of models to communicate reasoning in ways humans can understand, apply, and learn from. To investigate this, we introduce Knowledge Integration and Transfer Evaluation (KITE), a conceptual and experimental framework for Human-AI knowledge transfer capabilities and conduct the first large-scale human study (N=118) explicitly designed to measure it. In our two-phase setup, humans first ideate with an AI on problem-solving strategies, then independently implement solutions, isolating model explanations' influence on human understanding. Our findings reveal that although model benchmark performance correlates with collaborative outcomes, this relationship is notably inconsistent, featuring significant outliers, indicating that knowledge transfer requires dedicated optimization. Our analysis identifies behavioral and strategic factors mediating successful knowledge transfer. We release our code, dataset, and evaluation framework to support future work on communicatively aligned models.
Authors: Junjie Xing, Yeye He, Mengyu Zhou, Haoyu Dong, Shi Han, Lingjiao Chen, Dongmei Zhang, Surajit Chaudhuri, H. V. Jagadish
Abstract: Tables and table-based use cases play a crucial role in many important real-world applications, such as spreadsheets, databases, and computational notebooks, which traditionally require expert-level users like data engineers, data analysts, and database administrators to operate. Although LLMs have shown remarkable progress in working with tables (e.g., in spreadsheet and database copilot scenarios), comprehensive benchmarking of such capabilities remains limited. In contrast to an extensive and growing list of NLP benchmarks, evaluations of table-related tasks are scarce, and narrowly focus on tasks like NL-to-SQL and Table-QA, overlooking the broader spectrum of real-world tasks that professional users face. This gap limits our understanding and model progress in this important area. In this work, we introduce MMTU, a large-scale benchmark with over 30K questions across 25 real-world table tasks, designed to comprehensively evaluate models ability to understand, reason, and manipulate real tables at the expert-level. These tasks are drawn from decades' worth of computer science research on tabular data, with a focus on complex table tasks faced by professional users. We show that MMTU require a combination of skills -- including table understanding, reasoning, and coding -- that remain challenging for today's frontier models, where even frontier reasoning models like OpenAI o4-mini and DeepSeek R1 score only around 60%, suggesting significant room for improvement. We highlight key findings in our evaluation using MMTU and hope that this benchmark drives further advances in understanding and developing foundation models for structured data processing and analysis. Our code and data are available at https://github.com/MMTU-Benchmark/MMTU and https://huggingface.co/datasets/MMTU-benchmark/MMTU.
URLs: https://github.com/MMTU-Benchmark/MMTU, https://huggingface.co/datasets/MMTU-benchmark/MMTU.
Authors: Lianhao Zhou, Hongyi Ling, Keqiang Yan, Kaiji Zhao, Xiaoning Qian, Raymundo Arr\'oyave, Xiaofeng Qian, Shuiwang Ji
Abstract: We aim at designing language agents with greater autonomy for crystal materials discovery. While most of existing studies restrict the agents to perform specific tasks within predefined workflows, we aim to automate workflow planning given high-level goals and scientist intuition. To this end, we propose Materials Agent unifying Planning, Physics, and Scientists, known as MAPPS. MAPPS consists of a Workflow Planner, a Tool Code Generator, and a Scientific Mediator. The Workflow Planner uses large language models (LLMs) to generate structured and multi-step workflows. The Tool Code Generator synthesizes executable Python code for various tasks, including invoking a force field foundation model that encodes physics. The Scientific Mediator coordinates communications, facilitates scientist feedback, and ensures robustness through error reflection and recovery. By unifying planning, physics, and scientists, MAPPS enables flexible and reliable materials discovery with greater autonomy, achieving a five-fold improvement in stability, uniqueness, and novelty rates compared with prior generative models when evaluated on the MP-20 data. We provide extensive experiments across diverse tasks to show that MAPPS is a promising framework for autonomous materials discovery.
Authors: Kihyun Kim, Jiawei Zhang, Asuman Ozdaglar, Pablo A. Parrilo
Abstract: Conventional preference learning methods often prioritize opinions held more widely when aggregating preferences from multiple evaluators. This may result in policies that are biased in favor of some types of opinions or groups. The objective of this paper is to develop a novel preference learning framework capable of aligning aggregate opinions and policies proportionally with the true population distribution of evaluator preferences. Our approach infers the feasible set of evaluator population distributions directly from pairwise comparison data. Using these estimates, the algorithm constructs a policy that satisfies foundational axioms from social choice theory, namely monotonicity and Pareto efficiency, as well as our newly-introduced axioms of population-proportional representation and population-bounded robustness. We propose a soft-max relaxation method that smoothly trade-offs population-proportional representation with the selection of the Condorcet winner (which beats all other options in pairwise comparisons). Finally, we validate the effectiveness and scalability of our approach through experiments on both tabular recommendation tasks and large-scale language model alignment.
Authors: Gouki Minegishi, Hiroki Furuta, Takeshi Kojima, Yusuke Iwasawa, Yutaka Matsuo
Abstract: Recent large-scale reasoning models have achieved state-of-the-art performance on challenging mathematical benchmarks, yet the internal mechanisms underlying their success remain poorly understood. In this work, we introduce the notion of a reasoning graph, extracted by clustering hidden-state representations at each reasoning step, and systematically analyze three key graph-theoretic properties: cyclicity, diameter, and small-world index, across multiple tasks (GSM8K, MATH500, AIME 2024). Our findings reveal that distilled reasoning models (e.g., DeepSeek-R1-Distill-Qwen-32B) exhibit significantly more recurrent cycles (about 5 per sample), substantially larger graph diameters, and pronounced small-world characteristics (about 6x) compared to their base counterparts. Notably, these structural advantages grow with task difficulty and model capacity, with cycle detection peaking at the 14B scale and exploration diameter maximized in the 32B variant, correlating positively with accuracy. Furthermore, we show that supervised fine-tuning on an improved dataset systematically expands reasoning graph diameters in tandem with performance gains, offering concrete guidelines for dataset design aimed at boosting reasoning capabilities. By bridging theoretical insights into reasoning graph structures with practical recommendations for data construction, our work advances both the interpretability and the efficacy of large reasoning models.
Authors: Emil Biju, Shayan Talaei, Zhemin Huang, Mohammadreza Pourreza, Azalia Mirhoseini, Amin Saberi
Abstract: Large reasoning models (LRMs) excel at complex reasoning tasks but typically generate lengthy sequential chains-of-thought, resulting in long inference times before arriving at the final answer. To address this challenge, we introduce SPRINT, a novel post-training and inference-time framework designed to enable LRMs to dynamically identify and exploit opportunities for parallelization during their reasoning process. SPRINT incorporates an innovative data curation pipeline that reorganizes natural language reasoning trajectories into structured rounds of long-horizon planning and parallel execution. By fine-tuning LRMs on a small amount of such curated data, the models learn to dynamically identify independent subtasks within extended reasoning processes and effectively execute them in parallel. Through extensive evaluations, we show that the models fine-tuned with the SPRINT framework match the performance of reasoning models on complex domains such as mathematics while generating up to ~39% fewer sequential tokens on problems requiring more than 8000 output tokens. Finally, we observe consistent results transferred to two out-of-distribution tasks of GPQA and Countdown with up to 45% and 65% reduction in average sequential tokens for longer reasoning trajectories, while achieving the performance of the fine-tuned reasoning model.
Authors: Emmanuel Anaya Gonzalez, Sairam Vaidya, Kanghee Park, Ruyi Ji, Taylor Berg-Kirkpatrick, Loris D'Antoni
Abstract: Constrained decoding enables Language Models (LMs) to produce samples that provably satisfy hard constraints. However, existing constrained-decoding approaches often distort the underlying model distribution, a limitation that is especially problematic in applications like program fuzzing, where one wants to generate diverse and valid program inputs for testing purposes. We propose a new constrained sampling framework based on Markov Chain Monte Carlo (MCMC) that simultaneously satisfies three core desiderata: constraint satisfying (every sample satisfies the constraint), monotonically converging (the sampling process converges to the true conditional distribution), and efficient (high-quality samples emerge in few steps). Our method constructs a proposal distribution over valid outputs and applies a Metropolis-Hastings acceptance criterion based on the LM's likelihood, ensuring principled and efficient exploration of the constrained space. Empirically, our sampler outperforms existing methods on both synthetic benchmarks and real-world program fuzzing tasks.
Authors: Yesheng Zhang, Wenjian Sun, Yuheng Chen, Qingwei Liu, Qi Lin, Rui Zhang, Xu Zhao
Abstract: Complex interactions among agents present a significant challenge for autonomous driving in real-world scenarios. Recently, a promising approach has emerged, which formulates the interactions of agents as a level-k game framework. It effectively decouples agent policies by hierarchical game levels. However, this framework ignores both the varying driving complexities among agents and the dynamic changes in agent states across game levels, instead treating them uniformly. Consequently, redundant and error-prone computations are introduced into this framework. To tackle the issue, this paper proposes a metric, termed as Trajectory Entropy, to reveal the game status of agents within the level-k game framework. The key insight stems from recognizing the inherit relationship between agent policy uncertainty and the associated driving complexity. Specifically, Trajectory Entropy extracts statistical signals representing uncertainty from the multimodality trajectory prediction results of agents in the game. Then, the signal-to-noise ratio of this signal is utilized to quantify the game status of agents. Based on the proposed Trajectory Entropy, we refine the current level-k game framework through a simple gating mechanism, significantly improving overall accuracy while reducing computational costs. Our method is evaluated on the Waymo and nuPlan datasets, in terms of trajectory prediction, open-loop and closed-loop planning tasks. The results demonstrate the state-of-the-art performance of our method, with precision improved by up to 19.89% for prediction and up to 16.48% for planning.
Authors: Marilyn Bello, Rafael Bello, Maria-Matilde Garc\'ia, Ann Now\'e, Iv\'an Sevillano-Garc\'ia, Francisco Herrera
Abstract: The growing application of artificial intelligence in sensitive domains has intensified the demand for systems that are not only accurate but also explainable and trustworthy. Although explainable AI (XAI) methods have proliferated, many do not consider the diverse audiences that interact with AI systems: from developers and domain experts to end-users and society. This paper addresses how trust in AI is influenced by the design and delivery of explanations and proposes a multilevel framework that aligns explanations with the epistemic, contextual, and ethical expectations of different stakeholders. The framework consists of three layers: algorithmic and domain-based, human-centered, and social explainability. We highlight the emerging role of Large Language Models (LLMs) in enhancing the social layer by generating accessible, natural language explanations. Through illustrative case studies, we demonstrate how this approach facilitates technical fidelity, user engagement, and societal accountability, reframing XAI as a dynamic, trust-building process.
Authors: Yichi Zhang, Xin Luna Dong, Zhaojiang Lin, Andrea Madotto, Anuj Kumar, Babak Damavandi, Joyce Chai, Seungwhan Moon
Abstract: Recent advances in conversational AI have been substantial, but developing real-time systems for perceptual task guidance remains challenging. These systems must provide interactive, proactive assistance based on streaming visual inputs, yet their development is constrained by the costly and labor-intensive process of data collection and system evaluation. To address these limitations, we present a comprehensive framework with three key contributions. First, we introduce a novel data curation pipeline that synthesizes dialogues from annotated egocentric videos, resulting in \dataset, a large-scale synthetic dialogue dataset spanning multiple domains. Second, we develop a suite of automatic evaluation metrics, validated through extensive human studies. Third, we propose an end-to-end model that processes streaming video inputs to generate contextually appropriate responses, incorporating novel techniques for handling data imbalance and long-duration videos. This work lays the foundation for developing real-time, proactive AI assistants capable of guiding users through diverse tasks. Project page: https://pro-assist.github.io/
Authors: Katarzyna Kobalczyk, Mihaela van der Schaar
Abstract: Reward modelling from preference data is a crucial step in aligning large language models (LLMs) with human values, requiring robust generalisation to novel prompt-response pairs. In this work, we propose to frame this problem in a causal paradigm, providing the rich toolbox of causality to identify the persistent challenges, such as causal misidentification, preference heterogeneity, and confounding due to user-specific factors. Inheriting from the literature of causal inference, we identify key assumptions necessary for reliable generalisation and contrast them with common data collection practices. We illustrate failure modes of naive reward models and demonstrate how causally-inspired approaches can improve model robustness. Finally, we outline desiderata for future research and practices, advocating targeted interventions to address inherent limitations of observational data.
Authors: Qingbin Zeng, Ruotong Zhao, Jinzhu Mao, Haoyang Li, Fengli Xu, Yong Li
Abstract: Modeling urban crime is an important yet challenging task that requires understanding the subtle visual, social, and cultural cues embedded in urban environments. Previous work has predominantly focused on rule-based agent-based modeling (ABM) and deep learning methods. ABMs offer interpretability of internal mechanisms but exhibit limited predictive accuracy.In contrast, deep learning methods are often effective in prediction but are less interpretable and require extensive training data. Moreover, both lines of work lack the cognitive flexibility to adapt to changing environments. Leveraging the capabilities of large language models (LLMs), we propose CrimeMind, a novel LLM-driven ABM framework for simulating urban crime within a multi-modal urban context.A key innovation of our design is the integration of the Routine Activity Theory (RAT) into the agentic workflow of CrimeMind, enabling it to process rich multi-modal urban features and reason about criminal behavior.However, RAT requires LLM agents to infer subtle cues in evaluating environmental safety as part of assessing guardianship, which can be challenging for LLMs. To address this, we collect a small-scale human-annotated dataset and align CrimeMind's perception with human judgment via a training-free textual gradient method.Experiments across four major U.S. cities demonstrate that CrimeMind outperforms both traditional ABMs and deep learning baselines in crime hotspot prediction and spatial distribution accuracy, achieving up to a 24% improvement over the strongest baseline.Furthermore, we conduct counterfactual simulations of external incidents and policy interventions and it successfully captures the expected changes in crime patterns, demonstrating its ability to reflect counterfactual scenarios.Overall, CrimeMind enables fine-grained modeling of individual behaviors and facilitates evaluation of real-world interventions.
Authors: Kostis Michailidis, Dimos Tsouros, Tias Guns
Abstract: Combinatorial problems are present in a wide range of industries. Constraint Programming (CP) is a well-suited problem-solving paradigm, but its core process, namely constraint modelling, is a bottleneck for wider adoption. Aiming to alleviate this bottleneck, recent studies have explored using Large Language Models (LLMs) as modelling assistants, transforming combinatorial problem descriptions to executable constraint models, similar to coding assistants. However, the existing evaluation datasets for constraint modelling are often limited to small, homogeneous, or domain-specific instances, which do not capture the diversity of real-world scenarios. This work addresses this gap by introducing CP-Bench, a novel benchmark dataset that includes a diverse set of well-known combinatorial problem classes sourced from the CP community, structured explicitly for evaluating LLM-driven CP modelling. With this dataset, and given the variety of constraint modelling frameworks, we compare and evaluate the modelling capabilities of LLMs for three distinct constraint modelling systems, which vary in abstraction level and underlying syntax: the high-level MiniZinc language and Python-based CPMpy library, and the lower-level Python interface of the OR-Tools CP-SAT solver. In order to enhance the ability of LLMs to produce valid constraint models, we systematically evaluate the use of prompt-based and inference-time compute methods adapted from existing LLM-based code generation research. Our results underscore the modelling convenience provided by Python-based frameworks, as well as the effectiveness of documentation-rich system prompts, which, augmented with repeated sampling and self-verification, achieve further improvements, reaching up to 70\% accuracy on this new, highly challenging benchmark.
Authors: Ziyu Zhang, Peilan Xu, Yuetong Sun, Yuhui Shi, Wenjian Luo
Abstract: Large-scale itinerary planning is a variant of the traveling salesman problem, aiming to determine an optimal path that maximizes the collected points of interest (POIs) scores while minimizing travel time and cost, subject to travel duration constraints. This paper analyzes the decomposability of large-scale itinerary planning, proving that strict decomposability is difficult to satisfy, and introduces a weak decomposability definition based on a necessary condition, deriving the corresponding graph structures that fulfill this property. With decomposability guaranteed, we propose a novel multi-objective cooperative coevolutionary algorithm for large-scale itinerary planning, addressing the challenges of component imbalance and interactions. Specifically, we design a dynamic decomposition strategy based on the normalized fitness within each component, define optimization potential considering component scale and contribution, and develop a computational resource allocation strategy. Finally, we evaluate the proposed algorithm on a set of real-world datasets. Comparative experiments with state-of-the-art multi-objective itinerary planning algorithms demonstrate the superiority of our approach, with performance advantages increasing as the problem scale grows.
Authors: Jialu Zhang, Chu-Min Li, Sami Cherif, Shuolin Li, Zhifei Zheng
Abstract: The Maximum Satisfiability problem (MaxSAT) is a major optimization challenge with numerous practical applications. In recent MaxSAT evaluations, most MaxSAT solvers have adopted an ILP solver as part of their portfolios. This paper investigates the impact of Integer Linear Programming (ILP) preprocessing techniques on MaxSAT solving. Experimental results show that ILP preprocessing techniques help WMaxCDCL-OpenWbo1200, the winner of the MaxSAT evaluation 2024 in the unweighted track, solve 15 additional instances. Moreover, current state-of-the-art MaxSAT solvers heavily use an ILP solver in their portfolios, while our proposed approach reduces the need to call an ILP solver in a portfolio including WMaxCDCL or MaxCDCL.
Authors: Weizhi Zhang, Xinyang Zhang, Chenwei Zhang, Liangwei Yang, Jingbo Shang, Zhepei Wei, Henry Peng Zou, Zijie Huang, Zhengyang Wang, Yifan Gao, Xiaoman Pan, Lian Xiong, Jingguo Liu, Philip S. Yu, Xian Li
Abstract: Large Language Model (LLM) empowered agents have recently emerged as advanced paradigms that exhibit impressive capabilities in a wide range of domains and tasks. Despite their potential, current LLM agents often adopt a one-size-fits-all approach, lacking the flexibility to respond to users' varying needs and preferences. This limitation motivates us to develop PersonaAgent, the first personalized LLM agent framework designed to address versatile personalization tasks. Specifically, PersonaAgent integrates two complementary components - a personalized memory module that includes episodic and semantic memory mechanisms; a personalized action module that enables the agent to perform tool actions tailored to the user. At the core, the persona (defined as unique system prompt for each user) functions as an intermediary: it leverages insights from personalized memory to control agent actions, while the outcomes of these actions in turn refine the memory. Based on the framework, we propose a test-time user-preference alignment strategy that simulate the latest n interactions to optimize the persona prompt, ensuring real-time user preference alignment through textual loss feedback between simulated and ground-truth responses. Experimental evaluations demonstrate that PersonaAgent significantly outperforms other baseline methods by not only personalizing the action space effectively but also scaling during test-time real-world applications. These results underscore the feasibility and potential of our approach in delivering tailored, dynamic user experiences.
Authors: Jihwan Jeong, Xiaoyu Wang, Jingmin Wang, Scott Sanner, Pascal Poupart
Abstract: Offline reinforcement learning (RL) is crucial when online exploration is costly or unsafe but often struggles with high epistemic uncertainty due to limited data. Existing methods rely on fixed conservative policies, restricting adaptivity and generalization. To address this, we propose Reflect-then-Plan (RefPlan), a novel doubly Bayesian offline model-based (MB) planning approach. RefPlan unifies uncertainty modeling and MB planning by recasting planning as Bayesian posterior estimation. At deployment, it updates a belief over environment dynamics using real-time observations, incorporating uncertainty into MB planning via marginalization. Empirical results on standard benchmarks show that RefPlan significantly improves the performance of conservative offline RL policies. In particular, RefPlan maintains robust performance under high epistemic uncertainty and limited data, while demonstrating resilience to changing environment dynamics, improving the flexibility, generalizability, and robustness of offline-learned policies.
Authors: Chi Xie, Fangao Zeng, Yue Hu, Shuang Liang, Yichen Wei
Abstract: Unlike most previous HOI methods that focus on learning better human-object features, we propose a novel and complementary approach called category query learning. Such queries are explicitly associated to interaction categories, converted to image specific category representation via a transformer decoder, and learnt via an auxiliary image-level classification task. This idea is motivated by an earlier multi-label image classification method, but is for the first time applied for the challenging human-object interaction classification task. Our method is simple, general and effective. It is validated on three representative HOI baselines and achieves new state-of-the-art results on two benchmarks.
Authors: Linxuan He, Qing-Shan Jia, Ang Li, Hongyan Sang, Ling Wang, Jiwen Lu, Tao Zhang, Jie Zhou, Yi Zhang, Yisen Wang, Peng Wei, Zhongyuan Wang, Henry X. Liu, Shuo Feng
Abstract: Embodied AI systems, comprising AI models and physical plants, are increasingly prevalent across various applications. Due to the rarity of system failures, ensuring their safety in complex operating environments remains a major challenge, which severely hinders their large-scale deployment in safety-critical domains, such as autonomous vehicles, medical devices, and robotics. While achieving provable deterministic safety--verifying system safety across all possible scenarios--remains theoretically ideal, the rarity and complexity of corner cases make this approach impractical for scalable embodied AI systems. To address this challenge, we introduce provable probabilistic safety, which aims to ensure that the residual risk of large-scale deployment remains below a predefined threshold. Instead of attempting exhaustive safety proof across all corner cases, this paradigm establishes a probabilistic safety boundary on overall system performance, leveraging statistical methods to enhance feasibility and scalability. A well-defined probabilistic safety boundary enables embodied AI systems to be deployed at scale while allowing for continuous refinement of safety guarantees. Our work focuses on three core questions: what is provable probabilistic safety, how to prove the probabilistic safety, and how to achieve the provable probabilistic safety. By bridging the gap between theoretical safety assurance and practical deployment, our work offers a pathway toward safer, large-scale adoption of embodied AI systems in safety-critical applications.
Authors: Rukmal Weerawarana, Maxwell Braun
Abstract: This work establishes a rigorous theoretical foundation for analyzing deep learning systems by leveraging Infinite Time Turing Machines (ITTMs), which extend classical computation into transfinite ordinal steps. Using ITTMs, we reinterpret modern architectures like Transformers, revealing fundamental limitations in scalability, efficiency, and interpretability. Building on these insights, we propose the Universal State Machine (USM), a novel computational paradigm designed from first principles. The USM employs a dynamic, queryable computation graph that evolves in real time, enabling modular, interpretable, and resource-efficient computation. This framework not only overcomes the inefficiencies and rigidity of current models but also lays the groundwork for scalable, generalizable artificial intelligence systems.
Authors: Souradip Nath
Abstract: Multimodal Large Language Models (MLLMs) like GPT-4V are capable of reasoning across text and image modalities, showing promise in a variety of complex vision-language tasks. In this preliminary study, we investigate the out-of-the-box capabilities of GPT-4V in the domain of image forensics, specifically, in detecting image splicing manipulations. Without any task-specific fine-tuning, we evaluate GPT-4V using three prompting strategies: Zero-Shot (ZS), Few-Shot (FS), and Chain-of-Thought (CoT), applied over a curated subset of the CASIA v2.0 splicing dataset. Our results show that GPT-4V achieves competitive detection performance in zero-shot settings (more than 85% accuracy), with CoT prompting yielding the most balanced trade-off across authentic and spliced images. Qualitative analysis further reveals that the model not only detects low-level visual artifacts but also draws upon real-world contextual knowledge such as object scale, semantic consistency, and architectural facts, to identify implausible composites. While GPT-4V lags behind specialized state-of-the-art splicing detection models, its generalizability, interpretability, and encyclopedic reasoning highlight its potential as a flexible tool in image forensics.
Authors: Valentine Bernasconi, Gustavo Marfia
Abstract: Recent breakthroughs in generative AI have opened the door to new research perspectives in the domain of art and cultural heritage, where a large number of artifacts have been digitized. There is a need for innovation to ease the access and highlight the content of digital collections. Such innovations develop into creative explorations of the digital image in relation to its malleability and contemporary interpretation, in confrontation to the original historical object. Based on the concept of the autonomous image, we propose a new framework towards the production of self-explaining cultural artifacts using open-source large-language, face detection, text-to-speech and audio-to-animation models. The goal is to start from a digitized artwork and to automatically assemble a short video of the latter where the main character animates to explain its content. The whole process questions cultural biases encapsulated in large-language models, the potential of digital images and deepfakes of artworks for educational purposes, along with concerns of the field of art history regarding such creative diversions.
Authors: Zifan Wang, Christina Q. Knight, Jeremy Kritz, Willow E. Primack, Julian Michael
Abstract: Large Language Model (LLM) safeguards, which implement request refusals, have become a widely adopted mitigation strategy against misuse. At the intersection of adversarial machine learning and AI safety, safeguard red teaming has effectively identified critical vulnerabilities in state-of-the-art refusal-trained LLMs. However, in our view the many conference submissions on LLM red teaming do not, in aggregate, prioritize the right research problems. First, testing against clear product safety specifications should take a higher priority than abstract social biases or ethical principles. Second, red teaming should prioritize realistic threat models that represent the expanding risk landscape and what real attackers might do. Finally, we contend that system-level safety is a necessary step to move red teaming research forward, as AI models present new threats as well as affordances for threat mitigation (e.g., detection and banning of malicious users) once placed in a deployment context. Adopting these priorities will be necessary in order for red teaming research to adequately address the slate of new threats that rapid AI advances present today and will present in the very near future.
Authors: Seyed Moein Ayyoubzadeh, Kourosh Shahnazari, Mohammmadali Keshtparvar, MohammadAmin Fazli
Abstract: Training large language models (LLMs) requires vast amounts of high-quality data from institutions that face legal, privacy, and strategic constraints. Existing data procurement methods often rely on unverifiable trust or ignore heterogeneous provider costs. We introduce a mechanism-design framework for truthful, trust-minimized data sharing that ensures dominant-strategy incentive compatibility (DSIC), individual rationality, and weak budget balance, while rewarding data based on both quality and learning utility. We formalize a model where providers privately know their data cost and quality, and value arises solely from the data's contribution to model performance. Based on this, we propose the Quality-Weighted Marginal-Incentive Auction (Q-MIA), which ranks providers using a virtual cost metric and uses Myerson-style payments to ensure DSIC and budget feasibility. To support settings with limited liquidity or long-term incentives, we introduce the Marginal Utility Token (MUT), which allocates future rights based on marginal contributions. We unify these in Mixed-MIA, a hybrid mechanism balancing upfront payments and deferred rewards. All mechanisms support verifiable, privacy-preserving implementation. Theoretically and empirically, they outperform volume-based and trust-based baselines, eliciting higher-quality data under budget constraints while remaining robust to misreporting and collusion. This establishes a principled foundation for sustainable and fair data markets for future LLMs.
Authors: Francesco Panebianco, Mario D'Onghia, Stefano Zanero aand Michele Carminati
Abstract: Deep learning systems, critical in domains like autonomous vehicles, are vulnerable to adversarial examples (crafted inputs designed to mislead classifiers). This study investigates black-box adversarial attacks in computer vision. This is a realistic scenario, where attackers have query-only access to the target model. Three properties are introduced to evaluate attack feasibility: robustness to compression, stealthiness to automatic detection, and stealthiness to human inspection. State-of-the-Art methods tend to prioritize one criterion at the expense of others. We propose ECLIPSE, a novel attack method employing Gaussian blurring on sampled gradients and a local surrogate model. Comprehensive experiments on a public dataset highlight ECLIPSE's advantages, demonstrating its contribution to the trade-off between the three properties.
Authors: Zhuoxuan Cai, Jian Zhang, Xinbin Yuan, Pengtao Jiang, Wenxiang Chen, Bowen Tang, Lujian Yao, Qiyuan Wang, Jinwen Chen, Bo Li
Abstract: Recent studies demonstrate that multimodal large language models (MLLMs) can proficiently evaluate visual quality through interpretable assessments. However, existing approaches typically treat quality scoring and reasoning descriptions as separate tasks with disjoint optimization objectives, leading to a trade-off: models adept at quality reasoning descriptions struggle with precise score regression, while score-focused models lack interpretability. This limitation hinders the full potential of MLLMs in visual quality assessment, where accuracy and interpretability should be mutually reinforcing. To address this, we propose a unified two-stage training framework comprising a cold-start stage and a reinforcement learning-based fine-tuning stage. Specifically, in the first stage, we distill high-quality data from a teacher model through expert-designed prompts, initializing reasoning capabilities via cross-entropy loss supervision. In the second stage, we introduce a novel reward with Group Relative Policy Optimization (GRPO) to jointly optimize scoring accuracy and reasoning consistency. We designate the models derived from these two stages as Q-Ponder-CI and Q-Ponder. Extensive experiments show that Q-Ponder achieves state-of-the-art (SOTA) performance on quality score regression benchmarks, delivering up to 6.5% higher SRCC on cross-domain datasets. Furthermore, Q-Ponder significantly outperforms description-based SOTA models, including its teacher model Qwen-2.5-VL-72B, particularly in description accuracy and reasonableness, demonstrating the generalization potential over diverse tasks.
Authors: Lo Pang-Yun Ting, Chengshuai Zhao, Yu-Hua Zeng, Yuan Jee Lim, Kun-Ta Chuang
Abstract: Clinical note generation aims to automatically produce free-text summaries of a patient's condition and diagnostic process, with discharge instructions being a representative long-form example. While recent large language model (LLM)-based methods pre-trained on general clinical corpora show promise in clinical text generation, they fall short in producing long-form notes from limited patient information. In this paper, we propose R2AG, the first reinforced retriever for long-form discharge instruction generation based on pre-admission data. R2AG is trained with reinforcement learning to retrieve reasoning paths from a medical knowledge graph, providing explicit semantic guidance to the LLM. To bridge the information gap, we propose Group-Based Retriever Optimization (GRO) which improves retrieval quality with group-relative rewards, encouraging reasoning leaps for deeper inference by the LLM. Comprehensive experiments on the MIMIC-IV-Note dataset show that R2AG outperforms baselines in both clinical efficacy and natural language generation metrics. Further analysis reveals that R2AG fills semantic gaps in sparse input scenarios, and retrieved reasoning paths help LLMs avoid clinical misinterpretation by focusing on key evidence and following coherent reasoning.
Authors: Jaydip Sen, Saptarshi Sengupta. Subhasis Dasgupta
Abstract: This chapter explores advancements in decoding strategies for large language models (LLMs), focusing on enhancing the Locally Typical Sampling (LTS) algorithm. Traditional decoding methods, such as top-k and nucleus sampling, often struggle to balance fluency, diversity, and coherence in text generation. To address these challenges, Adaptive Semantic-Aware Typicality Sampling (ASTS) is proposed as an improved version of LTS, incorporating dynamic entropy thresholding, multi-objective scoring, and reward-penalty adjustments. ASTS ensures contextually coherent and diverse text generation while maintaining computational efficiency. Its performance is evaluated across multiple benchmarks, including story generation and abstractive summarization, using metrics such as perplexity, MAUVE, and diversity scores. Experimental results demonstrate that ASTS outperforms existing sampling techniques by reducing repetition, enhancing semantic alignment, and improving fluency.
Authors: Jerrin Bright, Zhibo Wang, Yuhao Chen, Sirisha Rambhatla, John Zelek, David Clausi
Abstract: Lack of input data for in-the-wild activities often results in low performance across various computer vision tasks. This challenge is particularly pronounced in uncommon human-centric domains like sports, where real-world data collection is complex and impractical. While synthetic datasets offer a promising alternative, existing approaches typically suffer from limited diversity in human appearance, motion, and scene composition due to their reliance on rigid asset libraries and hand-crafted rendering pipelines. To address this, we introduce Gen4D, a fully automated pipeline for generating diverse and photorealistic 4D human animations. Gen4D integrates expert-driven motion encoding, prompt-guided avatar generation using diffusion-based Gaussian splatting, and human-aware background synthesis to produce highly varied and lifelike human sequences. Based on Gen4D, we present SportPAL, a large-scale synthetic dataset spanning three sports: baseball, icehockey, and soccer. Together, Gen4D and SportPAL provide a scalable foundation for constructing synthetic datasets tailored to in-the-wild human-centric vision tasks, with no need for manual 3D modeling or scene design.
Authors: Israa A. Albadarneh, Bassam H. Hammo, Omar S. Al-Kadi
Abstract: Image captioning involves generating textual descriptions from input images, bridging the gap between computer vision and natural language processing. Recent advancements in transformer-based models have significantly improved caption generation by leveraging attention mechanisms for better scene understanding. While various surveys have explored deep learning-based approaches for image captioning, few have comprehensively analyzed attention-based transformer models across multiple languages. This survey reviews attention-based image captioning models, categorizing them into transformer-based, deep learning-based, and hybrid approaches. It explores benchmark datasets, discusses evaluation metrics such as BLEU, METEOR, CIDEr, and ROUGE, and highlights challenges in multilingual captioning. Additionally, this paper identifies key limitations in current models, including semantic inconsistencies, data scarcity in non-English languages, and limitations in reasoning ability. Finally, we outline future research directions, such as multimodal learning, real-time applications in AI-powered assistants, healthcare, and forensic analysis. This survey serves as a comprehensive reference for researchers aiming to advance the field of attention-based image captioning.
Authors: Lianming Huang, Haibo Hu, Yufei Cui, Jiacheng Zuo, Shangyu Wu, Nan Guan, Chun Jason Xue
Abstract: With the rapid advancement of autonomous driving, deploying Vision-Language Models (VLMs) to enhance perception and decision-making has become increasingly common. However, the real-time application of VLMs is hindered by high latency and computational overhead, limiting their effectiveness in time-critical driving scenarios. This challenge is particularly evident when VLMs exhibit over-inference, continuing to process unnecessary layers even after confident predictions have been reached. To address this inefficiency, we propose AD-EE, an Early Exit framework that incorporates domain characteristics of autonomous driving and leverages causal inference to identify optimal exit layers. We evaluate our method on large-scale real-world autonomous driving datasets, including Waymo and the corner-case-focused CODA, as well as on a real vehicle running the Autoware Universe platform. Extensive experiments across multiple VLMs show that our method significantly reduces latency, with maximum improvements reaching up to 57.58%, and enhances object detection accuracy, with maximum gains of up to 44%.
Authors: Patrik Czak\'o, G\'abor Kert\'esz, S\'andor Sz\'en\'asi
Abstract: We present SmoothRot, a novel post-training quantization technique to enhance the efficiency of 4-bit quantization in Large Language Models (LLMs). SmoothRot addresses the critical challenge of massive activation outliers, by integrating channel-wise scaling with Hadamard transformations. Our technique effectively transforms extreme outliers into quantization-friendly activations, significantly improving quantization accuracy. Experiments conducted on popular LLMs (LLaMA2 7B, LLaMA3.1 8B, and Mistral 7B) demonstrate that SmoothRot consistently reduces the performance gap between quantized and FP16 models by approximately 10-30\% across language generation and zero-shot reasoning tasks, without introducing additional inference latency. Code is available at https://github.com/czakop/smoothrot.
Authors: Mingfei Chen, Zijun Cui, Xiulong Liu, Jinlin Xiang, Caleb Zheng, Jingyuan Li, Eli Shlizerman
Abstract: 3D spatial reasoning in dynamic, audio-visual environments is a cornerstone of human cognition yet remains largely unexplored by existing Audio-Visual Large Language Models (AV-LLMs) and benchmarks, which predominantly focus on static or 2D scenes. We introduce SAVVY-Bench, the first benchmark for 3D spatial reasoning in dynamic scenes with synchronized spatial audio. SAVVY-Bench is comprised of thousands of relationships involving static and moving objects, and requires fine-grained temporal grounding, consistent 3D localization, and multi-modal annotation. To tackle this challenge, we propose SAVVY, a novel training-free reasoning pipeline that consists of two stages: (i) Egocentric Spatial Tracks Estimation, which leverages AV-LLMs as well as other audio-visual methods to track the trajectories of key objects related to the query using both visual and spatial audio cues, and (ii) Dynamic Global Map Construction, which aggregates multi-modal queried object trajectories and converts them into a unified global dynamic map. Using the constructed map, a final QA answer is obtained through a coordinate transformation that aligns the global map with the queried viewpoint. Empirical evaluation demonstrates that SAVVY substantially enhances performance of state-of-the-art AV-LLMs, setting a new standard and stage for approaching dynamic 3D spatial reasoning in AV-LLMs.
Authors: David Zagardo
Abstract: We revisit 1-bit gradient compression through the lens of mutual-information differential privacy (MI-DP). Building on signSGD, we propose FERRET--Fast and Effective Restricted Release for Ethical Training--which transmits at most one sign bit per parameter group with Bernoulli masking. Theory: We prove each fired group leaks at most ln 2 nats; after subsampling with rate s, the total privacy loss of G groups trained for T steps with firing probability p is epsilon = G * T * s * p * ln 2. Thus FERRET achieves MI-DP for epsilon in [0.1, 2] without additive noise. Practice: We evaluate three granularities--FERRET-MAX (finest), FERRET-EIGHTH (medium), and FERRET-2 (coarsest)--on five LLMs (137M-1.8B parameters) against DPSGD and Non-DP baselines. All methods trained for 1, 3, and 5 epochs. Utility: Across all settings, FERRET-MAX/EIGHTH beat DPSGD's perplexity. At epsilon=0.5, 5 epochs: FERRET-EIGHTH achieves 3.98 perplexity vs DPSGD's 11.61 (2.9x better), within 23% of Non-DP (3.25). Privacy: MI-AUC stays at chance for FERRET-MAX/EIGHTH (~0.51), matching DPSGD vs Non-DP's 0.76-0.99. FERRET-2 shows higher leakage (~0.55) due to lower headroom. Efficiency: Stricter budgets fire fewer signs, so FERRET uses 19-33% of DPSGD's training time and only 34-36% of Non-DP training time. Take-away: Sign-based MI-DP gets closer to achieving all three qualities of the privacy, utility, performance trilemma: FERRET trains up to 5x faster, achieves 3x lower perplexity compared to DPSGD and 1.2x greater than Non-DP, all while providing formal, mathematically provable privacy guarantees using zero additive noise. The results also show that, in certain instances, masked 1-bit updates can match non-private training utility while safeguarding data.
Authors: Kyungsoo Kim, Jeongsoo Ha, Yusung Kim
Abstract: Vision-based reinforcement learning requires efficient and robust representations of image-based observations, especially when the images contain distracting (task-irrelevant) elements such as shadows, clouds, and light. It becomes more important if those distractions are not exposed during training. We design a Self-Predictive Dynamics (SPD) method to extract task-relevant features efficiently, even in unseen observations after training. SPD uses weak and strong augmentations in parallel, and learns representations by predicting inverse and forward transitions across the two-way augmented versions. In a set of MuJoCo visual control tasks and an autonomous driving task (CARLA), SPD outperforms previous studies in complex observations, and significantly improves the generalization performance for unseen observations. Our code is available at https://github.com/unigary/SPD.
Authors: Jeongsoo Ha, Kyungsoo Kim, Yusung Kim
Abstract: Model-based reinforcement learning (MBRL) has been used to efficiently solve vision-based control tasks in highdimensional image observations. Although recent MBRL algorithms perform well in trained observations, they fail when faced with visual distractions in observations. These task-irrelevant distractions (e.g., clouds, shadows, and light) may be constantly present in real-world scenarios. In this study, we propose a novel self-supervised method, Dream to Generalize (Dr. G), for zero-shot MBRL. Dr. G trains its encoder and world model with dual contrastive learning which efficiently captures task-relevant features among multi-view data augmentations. We also introduce a recurrent state inverse dynamics model that helps the world model to better understand the temporal structure. The proposed methods can enhance the robustness of the world model against visual distractions. To evaluate the generalization performance, we first train Dr. G on simple backgrounds and then test it on complex natural video backgrounds in the DeepMind Control suite, and the randomizing environments in Robosuite. Dr. G yields a performance improvement of 117% and 14% over prior works, respectively. Our code is open-sourced and available at https://github.com/JeongsooHa/DrG.git
Authors: Fanqi Kong, Weiqin Zu, Xinyu Chen, Yaodong Yang, Song-Chun Zhu, Xue Feng
Abstract: The rich and multifaceted nature of human social interaction, encompassing multimodal cues, unobservable relations and mental states, and dynamical behavior, presents a formidable challenge for artificial intelligence. To advance research in this area, we introduce SIV-Bench, a novel video benchmark for rigorously evaluating the capabilities of Multimodal Large Language Models (MLLMs) across Social Scene Understanding (SSU), Social State Reasoning (SSR), and Social Dynamics Prediction (SDP). SIV-Bench features 2,792 video clips and 8,792 meticulously generated question-answer pairs derived from a human-LLM collaborative pipeline. It is originally collected from TikTok and YouTube, covering a wide range of video genres, presentation styles, and linguistic and cultural backgrounds. It also includes a dedicated setup for analyzing the impact of different textual cues-original on-screen text, added dialogue, or no text. Our comprehensive experiments on leading MLLMs reveal that while models adeptly handle SSU, they significantly struggle with SSR and SDP, where Relation Inference (RI) is an acute bottleneck, as further examined in our analysis. Our study also confirms the critical role of transcribed dialogue in aiding comprehension of complex social interactions. By systematically identifying current MLLMs' strengths and limitations, SIV-Bench offers crucial insights to steer the development of more socially intelligent AI. The dataset and code are available at https://kfq20.github.io/sivbench/.
Authors: Wenhao Wu, Fuhong Liu, Haoru Li, Zican Hu, Daoyi Dong, Chunlin Chen, Zhi Wang
Abstract: In-context reinforcement learning (ICRL) has emerged as a promising paradigm for adapting RL agents to downstream tasks through prompt conditioning. However, two notable challenges remain in fully harnessing in-context learning within RL domains: the intrinsic multi-modality of the state-action-reward data and the diverse, heterogeneous nature of decision tasks. To tackle these challenges, we propose \textbf{T2MIR} (\textbf{T}oken- and \textbf{T}ask-wise \textbf{M}oE for \textbf{I}n-context \textbf{R}L), an innovative framework that introduces architectural advances of mixture-of-experts (MoE) into transformer-based decision models. T2MIR substitutes the feedforward layer with two parallel layers: a token-wise MoE that captures distinct semantics of input tokens across multiple modalities, and a task-wise MoE that routes diverse tasks to specialized experts for managing a broad task distribution with alleviated gradient conflicts. To enhance task-wise routing, we introduce a contrastive learning method that maximizes the mutual information between the task and its router representation, enabling more precise capture of task-relevant information. The outputs of two MoE components are concatenated and fed into the next layer. Comprehensive experiments show that T2MIR significantly facilitates in-context learning capacity and outperforms various types of baselines. We bring the potential and promise of MoE to ICRL, offering a simple and scalable architectural enhancement to advance ICRL one step closer toward achievements in language and vision communities. Our code is available at https://github.com/NJU-RL/T2MIR.
Authors: Zishan Shu, Yufan Deng, Hongyu Zhang, Zhiwei Nie, Jie Chen
Abstract: Activity cliff prediction is a critical task in drug discovery and material design. Existing computational methods are limited to handling single binding targets, which restricts the applicability of these prediction models. In this paper, we present the Multi-Grained Target Perception network (MTPNet) to incorporate the prior knowledge of interactions between the molecules and their target proteins. Specifically, MTPNet is a unified framework for activity cliff prediction, which consists of two components: Macro-level Target Semantic (MTS) guidance and Micro-level Pocket Semantic (MPS) guidance. By this way, MTPNet dynamically optimizes molecular representations through multi-grained protein semantic conditions. To our knowledge, it is the first time to employ the receptor proteins as guiding information to effectively capture critical interaction details. Extensive experiments on 30 representative activity cliff datasets demonstrate that MTPNet significantly outperforms previous approaches, achieving an average RMSE improvement of 18.95% on top of several mainstream GNN architectures. Overall, MTPNet internalizes interaction patterns through conditional deep learning to achieve unified predictions of activity cliffs, helping to accelerate compound optimization and design. Codes are available at: https://github.com/ZishanShu/MTPNet.
Authors: Zhihao Tang, Chaozhuo Li, Litian Zhang, Xi Zhang
Abstract: Early prediction of Mild Cognitive Impairment (MCI) conversion is hampered by a trade-off between immediacy--making fast predictions from a single baseline sMRI--and accuracy--leveraging longitudinal scans to capture disease progression. We propose MCI-Diff, a diffusion-based framework that synthesizes clinically plausible future sMRI representations directly from baseline data, achieving both real-time risk assessment and high predictive performance. First, a multi-task sequence reconstruction strategy trains a shared denoising network on interpolation and extrapolation tasks to handle irregular follow-up sampling and learn robust latent trajectories. Second, an LLM-driven "linguistic compass" is introduced for clinical plausibility sampling: generated feature candidates are quantized, tokenized, and scored by a fine-tuned language model conditioned on expected structural biomarkers, guiding autoregressive generation toward realistic disease patterns. Experiments on ADNI and AIBL cohorts show that MCI-Diff outperforms state-of-the-art baselines, improving early conversion accuracy by 5-12%.
Authors: Ashwin Ramesh Babu, Sajad Mousavi, Vineet Gundecha, Sahand Ghorbanpour, Avisek Naug, Antonio Guillen, Ricardo Luna Gutierrez, Soumyendu Sarkar
Abstract: Vision-language models, which integrate computer vision and natural language processing capabilities, have demonstrated significant advancements in tasks such as image captioning and visual question and answering. However, similar to traditional models, they are susceptible to small perturbations, posing a challenge to their robustness, particularly in deployment scenarios. Evaluating the robustness of these models requires perturbations in both the vision and language modalities to learn their inter-modal dependencies. In this work, we train a generic surrogate model that can take both image and text as input and generate joint representation which is further used to generate adversarial perturbations for both the text and image modalities. This coordinated attack strategy is evaluated on the visual question and answering and visual reasoning datasets using various state-of-the-art vision-language models. Our results indicate that the proposed strategy outperforms other multi-modal attacks and single-modality attacks from the recent literature. Our results demonstrate their effectiveness in compromising the robustness of several state-of-the-art pre-trained multi-modal models such as instruct-BLIP, ViLT and others.
Authors: Mingjie Chen (Zhejiang University), Tiancheng Zhu (Huazhong University of Science,Technology), Mingxue Zhang (The State Key Laboratory of Blockchain,Data Security, Zhejiang University,Hangzhou High-Tech Zone), Yiling He (University College London), Minghao Lin (University of Southern California), Penghui Li (Columbia University), Kui Ren (The State Key Laboratory of Blockchain,Data Security, Zhejiang University)
Abstract: Binary code similarity detection (BCSD) serves as a fundamental technique for various software engineering tasks, e.g., vulnerability detection and classification. Attacks against such models have therefore drawn extensive attention, aiming at misleading the models to generate erroneous predictions. Prior works have explored various approaches to generating semantic-preserving variants, i.e., adversarial samples, to evaluate the robustness of the models against adversarial attacks. However, they have mainly relied on heuristic criteria or iterative greedy algorithms to locate salient code influencing the model output, failing to operate on a solid theoretical basis. Moreover, when processing programs with high complexities, such attacks tend to be time-consuming. In this work, we propose a novel optimization for adversarial attacks against BCSD models. In particular, we aim to improve the attacks in a challenging scenario, where the attack goal is to limit the model predictions to a specific range, i.e., the targeted attacks. Our attack leverages the superior capability of black-box, model-agnostic explainers in interpreting the model decision boundaries, thereby pinpointing the critical code snippet to apply semantic-preserving perturbations. The evaluation results demonstrate that compared with the state-of-the-art attacks, the proposed attacks achieve higher attack success rate in almost all scenarios, while also improving the efficiency and transferability. Our real-world case studies on vulnerability detection and classification further demonstrate the security implications of our attacks, highlighting the urgent need to further enhance the robustness of existing BCSD models.
Authors: Ashwin Ramesh Babu, Sajad Mousavi, Vineet Gundecha, Sahand Ghorbanpour, Avisek Naug, Antonio Guillen, Ricardo Luna Gutierrez, Soumyendu Sarkar
Abstract: Evaluating the robustness of Video classification models is very challenging, specifically when compared to image-based models. With their increased temporal dimension, there is a significant increase in complexity and computational cost. One of the key challenges is to keep the perturbations to a minimum to induce misclassification. In this work, we propose a multi-agent reinforcement learning approach (spatial and temporal) that cooperatively learns to identify the given video's sensitive spatial and temporal regions. The agents consider temporal coherence in generating fine perturbations, leading to a more effective and visually imperceptible attack. Our method outperforms the state-of-the-art solutions on the Lp metric and the average queries. Our method enables custom distortion types, making the robustness evaluation more relevant to the use case. We extensively evaluate 4 popular models for video action recognition on two popular datasets, HMDB-51 and UCF-101.
Authors: Yuxuan Yue, Zukang Xu, Zhihang Yuan, Dawei Yang, Jianglong Wu, Liqiang Nie
Abstract: Large Language Models (LLMs) face significant challenges in edge deployment due to their massive parameter scale. Vector Quantization (VQ), a clustering-based quantization method, serves as a prevalent solution to this issue for its extremely low-bit (even at 2-bit) and considerable accuracy. Since a vector is a quantity in mathematics and physics that has both direction and magnitude, existing VQ works typically quantize them in a coupled manner. However, we find that direction exhibits significantly greater sensitivity to quantization compared to the magnitude. For instance, when separately clustering the directions and magnitudes of weight vectors in LLaMA-2-7B, the accuracy drop of zero-shot tasks are 46.5\% and 2.3\%, respectively. This gap even increases with the reduction of clustering centers. Further, Euclidean distance, a common metric to access vector similarities in current VQ works, places greater emphasis on reducing the magnitude error. This property is contrary to the above finding, unavoidably leading to larger quantization errors. To these ends, this paper proposes Polar Coordinate Decoupled Vector Quantization (PCDVQ), an effective and efficient VQ framework consisting of two key modules: 1) Polar Coordinate Decoupling (PCD), which transforms vectors into their polar coordinate representations and perform independent quantization of the direction and magnitude parameters.2) Distribution Aligned Codebook Construction (DACC), which optimizes the direction and magnitude codebooks in accordance with the source distribution. Experimental results show that PCDVQ outperforms baseline methods at 2-bit level by at least 1.5\% zero-shot accuracy, establishing a novel paradigm for accurate and highly compressed LLMs.
Authors: Zikang Liu, Tongtian Yue, Yepeng Tang, Longteng Guo, Junxian Cai, Qingbin Liu, Xi Chen, Jing Liu
Abstract: Group Relative Policy Optimization (GRPO) enhances policy learning by computing gradients from relative comparisons among candidate outputs that share a common input prefix. Despite its effectiveness, GRPO introduces substantial computational overhead when processing long shared prefixes, which must be redundantly encoded for each group member. This inefficiency becomes a major scalability bottleneck in long-context learning scenarios. We propose Prefix Grouper, an efficient GRPO training algorithm that eliminates redundant prefix computation via a Shared-Prefix Forward strategy. In particular, by restructuring self-attention into two parts, our method enables the shared prefix to be encoded only once, while preserving full differentiability and compatibility with end-to-end training. We provide both theoretical and empirical evidence that Prefix Grouper is training-equivalent to standard GRPO: it yields identical forward outputs and backward gradients, ensuring that the optimization dynamics and final policy performance remain unchanged. Empirically, our experiments confirm that Prefix Grouper achieves consistent results while significantly reducing the computational cost of training, particularly in long-prefix scenarios. The proposed method is fully plug-and-play: it is compatible with existing GRPO-based architectures and can be seamlessly integrated into current training pipelines as a drop-in replacement, requiring no structural modifications and only minimal changes to input construction and attention computation. Prefix Grouper enables the use of larger group sizes under the same computational budget, thereby improving the scalability of GRPO to more complex tasks and larger models. Code is now available at https://github.com/johncaged/PrefixGrouper
Authors: Thomas Massena (IRIT, DTIPG - SNCF, UT3), L\'eo and\'eol (IMT, DTIPG - SNCF, UT3), Thibaut Boissin (IRIT, UT3), Franck Mamalet (IRIT, UT3), Corentin Friedrich (IRIT, UT3), Mathieu Serrurier (IRIT, UT3), S\'ebastien Gerchinovitz (IMT)
Abstract: Conformal Prediction (CP) has proven to be an effective post-hoc method for improving the trustworthiness of neural networks by providing prediction sets with finite-sample guarantees. However, under adversarial attacks, classical conformal guarantees do not hold anymore: this problem is addressed in the field of Robust Conformal Prediction. Several methods have been proposed to provide robust CP sets with guarantees under adversarial perturbations, but, for large scale problems, these sets are either too large or the methods are too computationally demanding to be deployed in real life scenarios. In this work, we propose a new method that leverages Lipschitz-bounded networks to precisely and efficiently estimate robust CP sets. When combined with a 1-Lipschitz robust network, we demonstrate that our lip-rcp method outperforms state-of-the-art results in both the size of the robust CP sets and computational efficiency in medium and large-scale scenarios such as ImageNet. Taking a different angle, we also study vanilla CP under attack, and derive new worst-case coverage bounds of vanilla CP sets, which are valid simultaneously for all adversarial attack levels. Our lip-rcp method makes this second approach as efficient as vanilla CP while also allowing robustness guarantees.
Authors: Manon Renault (IMT Atlantique), Hamoud Younes (IMT Atlantique), Hugo Tessier (IMT Atlantique), Ronan Le Roy (IMT Atlantique), Bastien Pasdeloup (IMT Atlantique), Mathieu L\'eonardon (IMT Atlantique)
Abstract: Package monitoring is an important topic in industrial applications, with significant implications for operational efficiency and ecological sustainability. In this study, we propose an approach that employs an embedded system, placed on reusable packages, to detect their state (on a Forklift, in a Truck, or in an undetermined location). We aim to design a system with a lifespan of several years, corresponding to the lifespan of reusable packages. Our analysis demonstrates that maximizing device lifespan requires minimizing wake time. We propose a pipeline that includes data processing, training, and evaluation of the deep learning model designed for imbalanced, multiclass time series data collected from an embedded sensor. The method uses a one-dimensional Convolutional Neural Network architecture to classify accelerometer data from the IoT device. Before training, two data augmentation techniques are tested to solve the imbalance problem of the dataset: the Synthetic Minority Oversampling TEchnique and the ADAptive SYNthetic sampling approach. After training, compression techniques are implemented to have a small model size. On the considered twoclass problem, the methodology yields a precision of 94.54% for the first class and 95.83% for the second class, while compression techniques reduce the model size by a factor of four. The trained model is deployed on the IoT device, where it operates with a power consumption of 316 mW during inference.
Authors: Julien Soul\'e, Jean-Paul Jamont, Michel Occello, Louis-Marie Traonouez, Paul Th\'eron
Abstract: Multi-Agent Systems (MAS) have been successfully applied in industry for their ability to address complex, distributed problems, especially in IoT-based systems. Their efficiency in achieving given objectives and meeting design requirements is strongly dependent on the MAS organization during the engineering process of an application-specific MAS. To design a MAS that can achieve given goals, available methods rely on the designer's knowledge of the deployment environment. However, high complexity and low readability in some deployment environments make the application of these methods to be costly or raise safety concerns. In order to ease the MAS organization design regarding those concerns, we introduce an original Assisted MAS Organization Engineering Approach (AOMEA). AOMEA relies on combining a Multi-Agent Reinforcement Learning (MARL) process with an organizational model to suggest relevant organizational specifications to help in MAS engineering.
Authors: Tongda Sun, Chen Yin, Huailiang Zheng, Yining Dong
Abstract: Health indicator (HI) plays a key role in degradation assessment and prognostics of rolling bearings. Although various HI construction methods have been investigated, most of them rely on expert knowledge for feature extraction and overlook capturing dynamic information hidden in sequential degradation processes, which limits the ability of the constructed HI for degradation trend representation and prognostics. To address these concerns, a novel dynamic HI that considers HI-level temporal dependence is constructed through an unsupervised framework. Specifically, a degradation feature learning module composed of a skip-connection-based autoencoder first maps raw signals to a representative degradation feature space (DFS) to automatically extract essential degradation features without the need for expert knowledge. Subsequently, in this DFS, a new HI-generating module embedded with an inner HI-prediction block is proposed for dynamic HI construction, where the temporal dependence between past and current HI states is guaranteed and modeled explicitly. On this basis, the dynamic HI captures the inherent dynamic contents of the degradation process, ensuring its effectiveness for degradation tendency modeling and future degradation prognostics. The experiment results on two bearing lifecycle datasets demonstrate that the proposed HI construction method outperforms comparison methods, and the constructed dynamic HI is superior for prognostic tasks.
Authors: Sho Takishita, Jay Gala, Abdelrahman Mohamed, Kentaro Inui, Yova Kementchedjhieva
Abstract: Many vision-language models (VLMs) that prove very effective at a range of multimodal task, build on CLIP-based vision encoders, which are known to have various limitations. We investigate the hypothesis that the strong language backbone in VLMs compensates for possibly weak visual features by contextualizing or enriching them. Using three CLIP-based VLMs, we perform controlled self-attention ablations on a carefully designed probing task. Our findings show that despite known limitations, CLIP visual representations offer ready-to-read semantic information to the language decoder. However, in scenarios of reduced contextualization in the visual representations, the language decoder can largely compensate for the deficiency and recover performance. This suggests a dynamic division of labor in VLMs and motivates future architectures that offload more visual processing to the language decoder.
Authors: Ludovic Arnould, Salim Khazem, Hugues Ali Mehenni
Abstract: Visual Language Models (VLMs) are now sufficiently advanced to support a broad range of applications, including answering complex visual questions, and are increasingly expected to interact with images in varied ways. To evaluate them, current benchmarks often focus on specific domains (e.g., reading charts), constructing datasets of annotated real images paired with pre-defined Multiple Choice Questions (MCQs) to report aggregate accuracy scores. However, such benchmarks entail high annotation costs, risk information leakage, and do not clarify whether failures stem from limitations in visual perception, reasoning, or general knowledge. We propose a new evaluation methodology, inspired by ophthalmologic diagnostics, leveraging procedural generation of synthetic images to obtain control over visual attributes and precisely reveal perception failures in VLMs. Specifically, we build collections of images with gradually more challenging variations in the content of interest (e.g., number of objects in a counting task) while holding other visual parameters constant. This diagnostic allows systematic stress testing and fine-grained failure analysis, shifting the focus from coarse benchmarking toward targeted and interpretable assessment of VLM capabilities. Our code is available at https://github.com/byoeval/BYO-EVAL.
Authors: Hao Jiang, Chuan Hu, Yukang Shi, Yuan He, Ke Wang, Xi Zhang, Zhipeng Zhang
Abstract: Vision-Language Models (VLMs) offer a promising approach to end-to-end autonomous driving due to their human-like reasoning capabilities. However, troublesome gaps remains between current VLMs and real-world autonomous driving applications. One major limitation is that existing datasets with loosely formatted language descriptions are not machine-friendly and may introduce redundancy. Additionally, high computational cost and massive scale of VLMs hinder the inference speed and real-world deployment. To bridge the gap, this paper introduces a structured and concise benchmark dataset, NuScenes-S, which is derived from the NuScenes dataset and contains machine-friendly structured representations. Moreover, we present FastDrive, a compact VLM baseline with 0.9B parameters. In contrast to existing VLMs with over 7B parameters and unstructured language processing(e.g., LLaVA-1.5), FastDrive understands structured and concise descriptions and generates machine-friendly driving decisions with high efficiency. Extensive experiments show that FastDrive achieves competitive performance on structured dataset, with approximately 20% accuracy improvement on decision-making tasks, while surpassing massive parameter baseline in inference speed with over 10x speedup. Additionally, ablation studies further focus on the impact of scene annotations (e.g., weather, time of day) on decision-making tasks, demonstrating their importance on decision-making tasks in autonomous driving.
Authors: Yiyu Lin, Yan Wang, You Zhou, Xinye Ni, Jiahui Wu, Sen Yang
Abstract: As a core mechanism of epigenetic regulation in eukaryotes, protein post-translational modifications (PTMs) require precise prediction to decipher dynamic life activity networks. To address the limitations of existing deep learning models in cross-modal feature fusion, domain generalization, and architectural optimization, this study proposes UniPTMs: the first unified framework for multi-type PTM prediction. The framework innovatively establishes a "Master-Slave" dual-path collaborative architecture: The master path dynamically integrates high-dimensional representations of protein sequences, structures, and evolutionary information through a Bidirectional Gated Cross-Attention (BGCA) module, while the slave path optimizes feature discrepancies and recalibration between structural and traditional features using a Low-Dimensional Fusion Network (LDFN). Complemented by a Multi-scale Adaptive convolutional Pyramid (MACP) for capturing local feature patterns and a Bidirectional Hierarchical Gated Fusion Network (BHGFN) enabling multi-level feature integration across paths, the framework employs a Hierarchical Dynamic Weighting Fusion (HDWF) mechanism to intelligently aggregate multimodal features. Enhanced by a novel Hierarchical Contrastive loss function for feature consistency optimization, UniPTMs demonstrates significant performance improvements (3.2%-11.4% MCC and 4.2%-14.3% AP increases) over state-of-the-art models across five modification types and transcends the Single-Type Prediction Paradigm. To strike a balance between model complexity and performance, we have also developed a lightweight variant named UniPTMs-mini.
Authors: Thanh Vinh Vo, Young Lee, Haozhe Ma, Chien Lu, Tze-Yun Leong
Abstract: Hidden confounders that influence both states and actions can bias policy learning in reinforcement learning (RL), leading to suboptimal or non-generalizable behavior. Most RL algorithms ignore this issue, learning policies from observational trajectories based solely on statistical associations rather than causal effects. We propose DoSAC (Do-Calculus Soft Actor-Critic with Backdoor Adjustment), a principled extension of the SAC algorithm that corrects for hidden confounding via causal intervention estimation. DoSAC estimates the interventional policy $\pi(a | \mathrm{do}(s))$ using the backdoor criterion, without requiring access to true confounders or causal labels. To achieve this, we introduce a learnable Backdoor Reconstructor that infers pseudo-past variables (previous state and action) from the current state to enable backdoor adjustment from observational data. This module is integrated into a soft actor-critic framework to compute both the interventional policy and its entropy. Empirical results on continuous control benchmarks show that DoSAC outperforms baselines under confounded settings, with improved robustness, generalization, and policy reliability.
Authors: Dror Ivry, Oran Nahum
Abstract: Large Language Models (LLMs) are increasingly powerful but remain vulnerable to prompt injection attacks, where malicious inputs cause the model to deviate from its intended instructions. This paper introduces Sentinel, a novel detection model, qualifire/prompt-injection-sentinel, based on the \answerdotai/ModernBERT-large architecture. By leveraging ModernBERT's advanced features and fine-tuning on an extensive and diverse dataset comprising a few open-source and private collections, Sentinel achieves state-of-the-art performance. This dataset amalgamates varied attack types, from role-playing and instruction hijacking to attempts to generate biased content, alongside a broad spectrum of benign instructions, with private datasets specifically targeting nuanced error correction and real-world misclassifications. On a comprehensive, unseen internal test set, Sentinel demonstrates an average accuracy of 0.987 and an F1-score of 0.980. Furthermore, when evaluated on public benchmarks, it consistently outperforms strong baselines like protectai/deberta-v3-base-prompt-injection-v2. This work details Sentinel's architecture, its meticulous dataset curation, its training methodology, and a thorough evaluation, highlighting its superior detection capabilities.
Authors: Andrei Mircea, Supriyo Chakraborty, Nima Chitsazan, Irina Rish, Ekaterina Lobacheva
Abstract: This work aims to understand how scaling improves language models, specifically in terms of training dynamics. We find that language models undergo loss deceleration early in training; an abrupt slowdown in the rate of loss improvement, resulting in piecewise linear behaviour of the loss curve in log-log space. Scaling up the model mitigates this transition by (1) decreasing the loss at which deceleration occurs, and (2) improving the log-log rate of loss improvement after deceleration. We attribute loss deceleration to a type of degenerate training dynamics we term zero-sum learning (ZSL). In ZSL, per-example gradients become systematically opposed, leading to destructive interference in per-example changes in loss. As a result, improving loss on one subset of examples degrades it on another, bottlenecking overall progress. Loss deceleration and ZSL provide new insights into the training dynamics underlying language model scaling laws, and could potentially be targeted directly to improve language models independent of scale. We make our code and artefacts available at: https://github.com/mirandrom/zsl
Authors: Seongmin Lee, Aeree Cho, Grace C. Kim, ShengYun Peng, Mansi Phute, Duen Horng Chau
Abstract: As large language models (LLMs) see wider real-world use, understanding and mitigating their unsafe behaviors is critical. Interpretation techniques can reveal causes of unsafe outputs and guide safety, but such connections with safety are often overlooked in prior surveys. We present the first survey that bridges this gap, introducing a unified framework that connects safety-focused interpretation methods, the safety enhancements they inform, and the tools that operationalize them. Our novel taxonomy, organized by LLM workflow stages, summarizes nearly 70 works at their intersections. We conclude with open challenges and future directions. This timely survey helps researchers and practitioners navigate key advancements for safer, more interpretable LLMs.
Authors: Hongbo Zhao, Fei Zhu, Rundong Wang, Gaofeng Meng, Zhaoxiang Zhang
Abstract: Recent Multimodal Large Language Models (MLLMs) excel in vision-language understanding but face challenges in adapting to dynamic real-world scenarios that require continuous integration of new knowledge and skills. While continual learning (CL) offers a potential solution, existing benchmarks and methods suffer from critical limitations. In this paper, we introduce MLLM-CL, a novel benchmark encompassing domain and ability continual learning, where the former focuses on independently and identically distributed (IID) evaluation across evolving mainstream domains, whereas the latter evaluates on non-IID scenarios with emerging model ability. Methodologically, we propose preventing catastrophic interference through parameter isolation, along with an MLLM-based routing mechanism. Extensive experiments demonstrate that our approach can integrate domain-specific knowledge and functional abilities with minimal forgetting, significantly outperforming existing methods.
Authors: Liang Zhang, Bingcong Li, Kiran Koshy Thekumparampil, Sewoong Oh, Michael Muehlebach, Niao He
Abstract: Zeroth-order methods are extensively used in machine learning applications where gradients are infeasible or expensive to compute, such as black-box attacks, reinforcement learning, and language model fine-tuning. Existing optimization theory focuses on convergence to an arbitrary stationary point, but less is known on the implicit regularization that provides a fine-grained characterization on which particular solutions are finally reached. We show that zeroth-order optimization with the standard two-point estimator favors solutions with small trace of Hessian, which is widely used in previous work to distinguish between sharp and flat minima. We further provide convergence rates of zeroth-order optimization to approximate flat minima for convex and sufficiently smooth functions, where flat minima are defined as the minimizers that achieve the smallest trace of Hessian among all optimal solutions. Experiments on binary classification tasks with convex losses and language model fine-tuning support our theoretical findings.
Authors: Mohammed Almutairi
Abstract: During the wake of the Covid-19 pandemic, the educational paradigm has experienced a major change from in person learning traditional to online platforms. The change of learning convention has impacted the teacher-student especially in non-verbal communication. The absent of non-verbal communication has led to a reliance on verbal feedback which diminished the efficacy of the educational experience. This paper explores the integration of sentiment analysis into learning management systems (LMS) to bridge the student-teacher's gap by offering an alternative approach to interpreting student feedback beyond its verbal context. The research involves data preparation, feature selection, and the development of a deep neural network model encompassing word embedding, LSTM, and attention mechanisms. This model is compared against a logistic regression baseline to evaluate its efficacy in understanding student feedback. The study aims to bridge the communication gap between instructors and students in online learning environments, offering insights into the emotional context of student feedback and ultimately improving the quality of online education.
Authors: Sima Noorani, Shayan Kiyani, George Pappas, Hamed Hassani
Abstract: Uncertainty quantification (UQ) is essential for safe deployment of generative AI models such as large language models (LLMs), especially in high stakes applications. Conformal prediction (CP) offers a principled uncertainty quantification framework, but classical methods focus on regression and classification, relying on geometric distances or softmax scores: tools that presuppose structured outputs. We depart from this paradigm by studying CP in a query only setting, where prediction sets must be constructed solely from finite queries to a black box generative model, introducing a new trade off between coverage, test time query budget, and informativeness. We introduce Conformal Prediction with Query Oracle (CPQ), a framework characterizing the optimal interplay between these objectives. Our finite sample algorithm is built on two core principles: one governs the optimal query policy, and the other defines the optimal mapping from queried samples to prediction sets. Remarkably, both are rooted in the classical missing mass problem in statistics. Specifically, the optimal query policy depends on the rate of decay, or the derivative, of the missing mass, for which we develop a novel estimator. Meanwhile, the optimal mapping hinges on the missing mass itself, which we estimate using Good Turing estimators. We then turn our focus to implementing our method for language models, where outputs are vast, variable, and often under specified. Fine grained experiments on three real world open ended tasks and two LLMs, show CPQ applicability to any black box LLM and highlight: (1) individual contribution of each principle to CPQ performance, and (2) CPQ ability to yield significantly more informative prediction sets than existing conformal methods for language uncertainty quantification.
Authors: Ya Jiang, Chuxiong Wu, Massieh Kordi Boroujeny, Brian Mark, Kai Zeng
Abstract: Watermarking for large language models (LLMs) offers a promising approach to identifying AI-generated text. Existing approaches, however, either compromise the distribution of original generated text by LLMs or are limited to embedding zero-bit information that only allows for watermark detection but ignores identification. We present StealthInk, a stealthy multi-bit watermarking scheme that preserves the original text distribution while enabling the embedding of provenance data, such as userID, TimeStamp, and modelID, within LLM-generated text. This enhances fast traceability without requiring access to the language model's API or prompts. We derive a lower bound on the number of tokens necessary for watermark detection at a fixed equal error rate, which provides insights on how to enhance the capacity. Comprehensive empirical evaluations across diverse tasks highlight the stealthiness, detectability, and resilience of StealthInk, establishing it as an effective solution for LLM watermarking applications.
Authors: Tiyasa Mitra, Ritika Borkar, Nidhi Bhatia, Ramon Matas, Shivam Raj, Dheevatsa Mudigere, Ritchie Zhao, Maximilian Golub, Arpan Dutta, Sailaja Madduri, Dharmesh Jani, Brian Pharris, Bita Darvish Rouhani
Abstract: As inference scales to multi-node deployments, disaggregation - splitting inference into distinct phases - offers a promising path to improving the throughput-interactivity Pareto frontier. Despite growing enthusiasm and a surge of open-source efforts, practical deployment of disaggregated serving remains limited due to the complexity of the optimization search space and system-level coordination. In this paper, we present the first systematic study of disaggregated inference at scale, evaluating hundreds of thousands of design points across diverse workloads and hardware configurations. We find that disaggregation is most effective for prefill-heavy traffic patterns and larger models. Our results highlight the critical role of dynamic rate matching and elastic scaling in achieving Pareto-optimal performance. Our findings offer actionable insights for efficient disaggregated deployments to navigate the trade-off between system throughput and interactivity.
Authors: Adrien Cort\'es, R\'emi Rehm, Victor Letzelter
Abstract: We introduce TimeMCL, a method leveraging the Multiple Choice Learning (MCL) paradigm to forecast multiple plausible time series futures. Our approach employs a neural network with multiple heads and utilizes the Winner-Takes-All (WTA) loss to promote diversity among predictions. MCL has recently gained attention due to its simplicity and ability to address ill-posed and ambiguous tasks. We propose an adaptation of this framework for time-series forecasting, presenting it as an efficient method to predict diverse futures, which we relate to its implicit quantization objective. We provide insights into our approach using synthetic data and evaluate it on real-world time series, demonstrating its promising performance at a light computational cost.
Authors: Boyuan Deng, Luca Rossini, Jin Wang, Weijie Wang, Nikolaos Tsagarakis
Abstract: Adaptive recovery from fall incidents are essential skills for the practical deployment of wheeled-legged robots, which uniquely combine the agility of legs with the speed of wheels for rapid recovery. However, traditional methods relying on preplanned recovery motions, simplified dynamics or sparse rewards often fail to produce robust recovery policies. This paper presents a learning-based framework integrating Episode-based Dynamic Reward Shaping and curriculum learning, which dynamically balances exploration of diverse recovery maneuvers with precise posture refinement. An asymmetric actor-critic architecture accelerates training by leveraging privileged information in simulation, while noise-injected observations enhance robustness against uncertainties. We further demonstrate that synergistic wheel-leg coordination reduces joint torque consumption by 15.8% and 26.2% and improves stabilization through energy transfer mechanisms. Extensive evaluations on two distinct quadruped platforms achieve recovery success rates up to 99.1% and 97.8% without platform-specific tuning. The supplementary material is available at https://boyuandeng.github.io/L2R-WheelLegCoordination/
URLs: https://boyuandeng.github.io/L2R-WheelLegCoordination/
Authors: Zikui Cai, Andrew Wang, Anirudh Satheesh, Ankit Nakhawa, Hyunwoo Jae, Keenan Powell, Minghui Liu, Neel Jay, Sungbin Oh, Xiyao Wang, Yongyuan Liang, Tom Goldstein, Furong Huang
Abstract: Despite rapid advances in vision-language models (VLMs), current benchmarks for multimodal reasoning fall short in three key dimensions. First, they overwhelmingly rely on static images, failing to capture the temporal complexity of real-world environments. Second, they narrowly focus on mathematical problem-solving, neglecting the broader spectrum of reasoning skills -- including abstract, physical, planning, spatial, and temporal capabilities -- required for robust multimodal intelligence. Third, many benchmarks quickly saturate, offering limited headroom for diagnosing failure modes or measuring continued progress. We introduce MORSE-500 (Multimodal Reasoning Stress-test Environment), a video benchmark composed of 500 fully scripted clips with embedded questions spanning six complementary reasoning categories. Each instance is programmatically generated using deterministic Python scripts (via Manim, Matplotlib, MoviePy), generative video models, and curated real footage. This script-driven design allows fine-grained control over visual complexity, distractor density, and temporal dynamics -- enabling difficulty to be scaled systematically as models improve. Unlike static benchmarks that become obsolete once saturated, MORSE-500 is built to evolve: its controllable generation pipeline supports the creation of arbitrarily challenging new instances, making it ideally suited for stress-testing next-generation models. Initial experiments with state-of-the-art systems -- including various Gemini 2.5 Pro and OpenAI o3 which represent the strongest available at the time, alongside strong open-source models -- reveal substantial performance gaps across all categories, with particularly large deficits in abstract and planning tasks. We release the full dataset, generation scripts, and evaluation harness to support transparent, reproducible, and forward-looking multimodal reasoning research.
Authors: Feliks Ba\'nka, Jaros{\l}aw A. Chudziak
Abstract: Accurate option pricing is essential for effective trading and risk management in financial markets, yet it remains challenging due to market volatility and the limitations of traditional models like Black-Scholes. In this paper, we investigate the application of the Informer neural network for option pricing, leveraging its ability to capture long-term dependencies and dynamically adjust to market fluctuations. This research contributes to the field of financial forecasting by introducing Informer's efficient architecture to enhance prediction accuracy and provide a more adaptable and resilient framework compared to existing methods. Our results demonstrate that Informer outperforms traditional approaches in option pricing, advancing the capabilities of data-driven financial forecasting in this domain.
Authors: Chenhui Deng, Yun-Da Tsai, Guan-Ting Liu, Zhongzhi Yu, Haoxing Ren
Abstract: Recent advances in large language models (LLMs) have enabled near-human performance on software coding benchmarks, but their effectiveness in RTL code generation remains limited due to the scarcity of high-quality training data. While prior efforts have fine-tuned LLMs for RTL tasks, they do not fundamentally overcome the data bottleneck and lack support for test-time scaling due to their non-reasoning nature. In this work, we introduce ScaleRTL, the first reasoning LLM for RTL coding that scales up both high-quality reasoning data and test-time compute. Specifically, we curate a diverse set of long chain-of-thought reasoning traces averaging 56K tokens each, resulting in a dataset of 3.5B tokens that captures rich RTL knowledge. Fine-tuning a general-purpose reasoning model on this corpus yields ScaleRTL that is capable of deep RTL reasoning. Subsequently, we further enhance the performance of ScaleRTL through a novel test-time scaling strategy that extends the reasoning process via iteratively reflecting on and self-correcting previous reasoning steps. Experimental results show that ScaleRTL achieves state-of-the-art performance on VerilogEval and RTLLM, outperforming 18 competitive baselines by up to 18.4% on VerilogEval and 12.7% on RTLLM.
Authors: Arian Raje, Baris Askin, Divyansh Jhunjhunwala, Gauri Joshi
Abstract: Large language models (LLMs) have not yet effectively leveraged the vast amounts of edge-device data, and federated learning (FL) offers a promising paradigm to collaboratively fine-tune LLMs without transferring private edge data to the cloud. To operate within the computation and communication constraints of edge devices, recent literature on federated fine-tuning of LLMs proposes the use of low-rank adaptation (LoRA) and similar parameter-efficient methods. However, LoRA-based methods suffer from accuracy degradation in FL settings, primarily because of data and computational heterogeneity across clients. We propose \textsc{Ravan}, an adaptive multi-head LoRA method that balances parameter efficiency and model expressivity by reparameterizing the weight updates as the sum of multiple LoRA heads $s_i\textbf{B}_i\textbf{H}_i\textbf{A}_i$ in which only the core matrices $\textbf{H}_i$ and their lightweight scaling factors $s_i$ are trained. These trainable scaling factors let the optimization focus on the most useful heads, recovering a higher-rank approximation of the full update without increasing the number of communicated parameters since clients upload $s_i\textbf{H}_i$ directly. Experiments on vision and language benchmarks show that \textsc{Ravan} improves test accuracy by 2-8\% over prior parameter-efficient baselines, making it a robust and scalable solution for federated fine-tuning of LLMs.
Authors: Saptarshi Nath, Christos Peridis, Eseoghene Benjamin, Xinran Liu, Soheil Kolouri, Peter Kinnell, Zexin Li, Cong Liu, Shirin Dora, Andrea Soltoggio
Abstract: Agentic AI has gained significant interest as a research paradigm focused on autonomy, self-directed learning, and long-term reliability of decision making. Real-world agentic systems operate in decentralized settings on a large set of tasks or data distributions with constraints such as limited bandwidth, asynchronous execution, and the absence of a centralized model or even common objectives. We posit that exploiting previously learned skills, task similarities, and communication capabilities in a collective of agentic AI are challenging but essential elements to enabling scalability, open-endedness, and beneficial collaborative learning dynamics. In this paper, we introduce Modular Sharing and Composition in Collective Learning (MOSAIC), an agentic algorithm that allows multiple agents to independently solve different tasks while also identifying, sharing, and reusing useful machine-learned knowledge, without coordination, synchronization, or centralized control. MOSAIC combines three mechanisms: (1) modular policy composition via neural network masks, (2) cosine similarity estimation using Wasserstein embeddings for knowledge selection, and (3) asynchronous communication and policy integration. Results on a set of RL benchmarks show that MOSAIC has a greater sample efficiency than isolated learners, i.e., it learns significantly faster, and in some cases, finds solutions to tasks that cannot be solved by isolated learners. The collaborative learning and sharing dynamics are also observed to result in the emergence of ideal curricula of tasks, from easy to hard. These findings support the case for collaborative learning in agentic systems to achieve better and continuously evolving performance both at the individual and collective levels.
Authors: Azza Abouzied, Firoj Alam, Raian Ali, Paolo Papotti
Abstract: Misinformation and disinformation pose significant risks globally, with the Arab region facing unique vulnerabilities due to geopolitical instabilities, linguistic diversity, and cultural nuances. We explore these challenges through the key facets of combating misinformation: detection, tracking, mitigation and community-engagement. We shed light on how connecting with grass-roots fact-checking organizations, understanding cultural norms, promoting social correction, and creating strong collaborative information networks can create opportunities for a more resilient information ecosystem in the Arab world.
Authors: Nien-Shao Wang, Duygu Nur Yaldiz, Yavuz Faruk Bakman, Sai Praneeth Karimireddy
Abstract: Conformal prediction is widely used to equip black-box machine learning models with uncertainty quantification enjoying formal coverage guarantees. However, these guarantees typically break down in the presence of distribution shifts, where the data distribution at test time differs from the training (or calibration-time) distribution. In this work, we address subpopulation shifts, where the test environment exhibits an unknown and differing mixture of subpopulations compared to the calibration data. We propose new methods that provably adapt conformal prediction to such shifts, ensuring valid coverage without requiring explicit knowledge of subpopulation structure. Our algorithms scale to high-dimensional settings and perform effectively in realistic machine learning tasks. Extensive experiments on vision (with vision transformers) and language (with large language models) benchmarks demonstrate that our methods reliably maintain coverage and controls risk in scenarios where standard conformal prediction fails.
Authors: Isha Puri, Amit Dhurandhar, Tejaswini Pedapati, Kartikeyan Shanmugam, Dennis Wei, Kush R. Varshney
Abstract: In recent years there has been a considerable amount of research on local post hoc explanations for neural networks. However, work on building interpretable neural architectures has been relatively sparse. In this paper, we present a novel neural architecture, CoFrNet, inspired by the form of continued fractions which are known to have many attractive properties in number theory, such as fast convergence of approximations to real numbers. We show that CoFrNets can be efficiently trained as well as interpreted leveraging their particular functional form. Moreover, we prove that such architectures are universal approximators based on a proof strategy that is different than the typical strategy used to prove universal approximation results for neural networks based on infinite width (or depth), which is likely to be of independent interest. We experiment on nonlinear synthetic functions and are able to accurately model as well as estimate feature attributions and even higher order terms in some cases, which is a testament to the representational power as well as interpretability of such architectures. To further showcase the power of CoFrNets, we experiment on seven real datasets spanning tabular, text and image modalities, and show that they are either comparable or significantly better than other interpretable models and multilayer perceptrons, sometimes approaching the accuracies of state-of-the-art models.
Authors: David Palzer, Matthew Maciejewski, Eric Fosler-Lussier
Abstract: In recent years, end-to-end approaches have made notable progress in addressing the challenge of speaker diarization, which involves segmenting and identifying speakers in multi-talker recordings. One such approach, Encoder-Decoder Attractors (EDA), has been proposed to handle variable speaker counts as well as better guide the network during training. In this study, we extend the attractor paradigm by moving beyond direct speaker modeling and instead focus on representing more detailed `speaker attributes' through a multi-stage process of intermediate representations. Additionally, we enhance the architecture by replacing transformers with conformers, a convolution-augmented transformer, to model local dependencies. Experiments demonstrate improved diarization performance on the CALLHOME dataset.
Authors: Jes Frellsen, Maher M. Kassem, Tone Bengtsen, Lars Olsen, Kresten Lindorff-Larsen, Jesper Ferkinghoff-Borg, Wouter Boomsma
Abstract: Inverse folding models have proven to be highly effective zero-shot predictors of protein stability. Despite this success, the link between the amino acid preferences of an inverse folding model and the free-energy considerations underlying thermodynamic stability remains incompletely understood. A better understanding would be of interest not only from a theoretical perspective, but also potentially provide the basis for stronger zero-shot stability prediction. In this paper, we take steps to clarify the free-energy foundations of inverse folding models. Our derivation reveals the standard practice of likelihood ratios as a simplistic approximation and suggests several paths towards better estimates of the relative stability. We empirically assess these approaches and demonstrate that considerable gains in zero-shot performance can be achieved with fairly simple means.
Authors: Michael J Ryan, Omar Shaikh, Aditri Bhagirath, Daniel Frees, William Held, Diyi Yang
Abstract: Recent calls for pluralistic alignment of Large Language Models (LLMs) encourage adapting models to diverse user preferences. However, most prior work on personalized reward models heavily rely on additional identity information, such as demographic details or a predefined set of preference categories. To this end, we introduce SynthesizeMe, an approach to inducing synthetic user personas from user interactions for personalized reward modeling. SynthesizeMe first generates and verifies reasoning to explain user preferences, then induces synthetic user personas from that reasoning, and finally filters to informative prior user interactions in order to build personalized prompts for a particular user. We show that using SynthesizeMe induced prompts improves personalized LLM-as-a-judge accuracy by 4.4% on Chatbot Arena. Combining SynthesizeMe derived prompts with a reward model achieves top performance on PersonalRewardBench: a new curation of user-stratified interactions with chatbots collected from 854 users of Chatbot Arena and PRISM.
Authors: Julia Barnett, Kimon Kieslich, Jasmine Sinchai, Nicholas Diakopoulos
Abstract: Scenario building is an established method to anticipate the future of emerging technologies. Its primary goal is to use narratives to map future trajectories of technology development and sociotechnical adoption. Following this process, risks and benefits can be identified early on, and strategies can be developed that strive for desirable futures. In recent years, computer science has adopted this method and applied it to various technologies, including Artificial Intelligence (AI). Because computing technologies play such an important role in shaping modern societies, it is worth exploring how scenarios are being used as an anticipatory tool in the field -- and what possible traditional uses of scenarios are not yet covered but have the potential to enrich the field. We address this gap by conducting a systematic literature review on the use of scenario building methods in computer science over the last decade (n = 59). We guide the review along two main questions. First, we aim to uncover how scenarios are used in computing literature, focusing especially on the rationale for why scenarios are used. Second, in following the potential of scenario building to enhance inclusivity in research, we dive deeper into the participatory element of the existing scenario building literature in computer science.
Authors: Ruipeng Zhang, Ya-Chien Chang, Sicun Gao
Abstract: The Maximum Entropy Reinforcement Learning (MaxEnt RL) framework is a leading approach for achieving efficient learning and robust performance across many RL tasks. However, MaxEnt methods have also been shown to struggle with performance-critical control problems in practice, where non-MaxEnt algorithms can successfully learn. In this work, we analyze how the trade-off between robustness and optimality affects the performance of MaxEnt algorithms in complex control tasks: while entropy maximization enhances exploration and robustness, it can also mislead policy optimization, leading to failure in tasks that require precise, low-entropy policies. Through experiments on a variety of control problems, we concretely demonstrate this misleading effect. Our analysis leads to better understanding of how to balance reward design and entropy maximization in challenging control problems.
Authors: Antonia van Betteray, Matthias Rottmann, Karsten Kahl
Abstract: The singular values of convolutional mappings encode interesting spectral properties, which can be used, e.g., to improve generalization and robustness of convolutional neural networks as well as to facilitate model compression. However, the computation of singular values is typically very resource-intensive. The naive approach involves unrolling the convolutional mapping along the input and channel dimensions into a large and sparse two-dimensional matrix, making the exact calculation of all singular values infeasible due to hardware limitations. In particular, this is true for matrices that represent convolutional mappings with large inputs and a high number of channels. Existing efficient methods leverage the Fast Fourier transformation (FFT) to transform convolutional mappings into the frequency domain, enabling the computation of singular values for matrices representing convolutions with larger input and channel dimensions. For a constant number of channels in a given convolution, an FFT can compute N singular values in O(N log N) complexity. In this work, we propose an approach of complexity O(N) based on local Fourier analysis, which additionally exploits the shift invariance of convolutional operators. We provide a theoretical analysis of our algorithm's runtime and validate its efficiency through numerical experiments. Our results demonstrate that our proposed method is scalable and offers a practical solution to calculate the entire set of singular values - along with the corresponding singular vectors if needed - for high-dimensional convolutional mappings.
Authors: Tianyi Zhang, Shidong Pan, Zejun Zhang, Zhenchang Xing, Xiaoyu Sun
Abstract: Infrastructure-as-Code (IaC) generation holds significant promise for automating cloud infrastructure provisioning. Recent advances in Large Language Models (LLMs) present a promising opportunity to democratize IaC development by generating deployable infrastructure templates from natural language descriptions, but current evaluation focuses on syntactic correctness while ignoring deployability, the fatal measure of IaC template utility. We address this gap through two contributions: (1) IaCGen, an LLM-based deployability-centric framework that uses iterative feedback mechanism to generate IaC templates, and (2) DPIaC-Eval, a deployability-centric IaC template benchmark consists of 153 real-world scenarios that can evaluate syntax, deployment, user intent, and security. Our evaluation reveals that state-of-the-art LLMs initially performed poorly, with Claude-3.5 and Claude-3.7 achieving only 30.2% and 26.8% deployment success on the first attempt respectively. However, IaCGen transforms this performance dramatically: all evaluated models reach over 90% passItr@25, with Claude-3.5 and Claude-3.7 achieving 98% success rate. Despite these improvements, critical challenges remain in user intent alignment (25.2% accuracy) and security compliance (8.4% pass rate), highlighting areas requiring continued research. Our work provides the first comprehensive assessment of deployability-centric IaC template generation and establishes a foundation for future research.
Authors: Jiri Navratil, Jarret Ross, Payel Das, Youssef Mroueh, Samuel C Hoffman, Vijil Chenthamarakshan, Brian Belgodere
Abstract: The ability to design molecules while preserving similarity to a target molecule and/or property is crucial for various applications in drug discovery, chemical design, and biology. We introduce in this paper an efficient training-free method for navigating and sampling from the molecular space with a generative Chemical Language Model (CLM), while using the molecular similarity to the target as a guide. Our method leverages the contextual representations learned from the CLM itself to estimate the molecular similarity, which is then used to adjust the autoregressive sampling strategy of the CLM. At each step of the decoding process, the method tracks the distance of the current generations from the target and updates the logits to encourage the preservation of similarity in generations. We implement the method using a recently proposed $\sim$47M parameter SMILES-based CLM, GP-MoLFormer, and therefore refer to the method as GP-MoLFormer-Sim, which enables a test-time update of the deep generative policy to reflect the contextual similarity to a set of guide molecules. The method is further integrated into a genetic algorithm (GA) and tested on a set of standard molecular optimization benchmarks involving property optimization, molecular rediscovery, and structure-based drug design. Results show that, GP-MoLFormer-Sim, combined with GA (GP-MoLFormer-Sim+GA) outperforms existing training-free baseline methods, when the oracle remains black-box. The findings in this work are a step forward in understanding and guiding the generative mechanisms of CLMs.
Authors: Saeed Hedayatian, Stefanos Nikolaidis
Abstract: Quality-Diversity (QD) algorithms have shown remarkable success in discovering diverse, high-performing solutions, but rely heavily on hand-crafted behavioral descriptors that constrain exploration to predefined notions of diversity. Leveraging the equivalence between policies and occupancy measures, we present a theoretically grounded approach to automatically generate behavioral descriptors by embedding the occupancy measures of policies in Markov Decision Processes. Our method, AutoQD, leverages random Fourier features to approximate the Maximum Mean Discrepancy (MMD) between policy occupancy measures, creating embeddings whose distances reflect meaningful behavioral differences. A low-dimensional projection of these embeddings that captures the most behaviorally significant dimensions is then used as behavioral descriptors for off-the-shelf QD methods. We prove that our embeddings converge to true MMD distances between occupancy measures as the number of sampled trajectories and embedding dimensions increase. Through experiments in multiple continuous control tasks we demonstrate AutoQD's ability in discovering diverse policies without predefined behavioral descriptors, presenting a well-motivated alternative to prior methods in unsupervised Reinforcement Learning and QD optimization. Our approach opens new possibilities for open-ended learning and automated behavior discovery in sequential decision making settings without requiring domain-specific knowledge.
Authors: Markelle Kelly, Alex Boyd, Sam Showalter, Mark Steyvers, Padhraic Smyth
Abstract: Applications of machine learning often involve making predictions based on both model outputs and the opinions of human experts. In this context, we investigate the problem of querying experts for class label predictions, using as few human queries as possible, and leveraging the class probability estimates of pre-trained classifiers. We develop a general Bayesian framework for this problem, modeling expert correlation via a joint latent representation, enabling simulation-based inference about the utility of additional expert queries, as well as inference of posterior distributions over unobserved expert labels. We apply our approach to two real-world medical classification problems, as well as to CIFAR-10H and ImageNet-16H, demonstrating substantial reductions relative to baselines in the cost of querying human experts while maintaining high prediction accuracy.
Authors: Markian Mandzak, Elvira Yang, Anna Zapaishchykova, Yu-Hui Chen, Lucas Heilbroner, John Zielke, Divyanshu Tak, Reza Mojahed-Yazdi, Francesca Romana Mussa, Zezhong Ye, Sridhar Vajapeyam, Viviana Benitez, Ralph Salloum, Susan N. Chi, Houman Sotoudeh, Jakob Seidlitz, Sabine Mueller, Hugo J. W. L. Aerts, Tina Y. Poussaint, Benjamin H. Kann
Abstract: Extracranial tissues visible on brain magnetic resonance imaging (MRI) may hold significant value for characterizing health conditions and clinical decision-making, yet they are rarely quantified. Current tools have not been widely validated, particularly in settings of developing brains or underlying pathology. We present TissUnet, a deep learning model that segments skull bone, subcutaneous fat, and muscle from routine three-dimensional T1-weighted MRI, with or without contrast enhancement. The model was trained on 155 paired MRI-computed tomography (CT) scans and validated across nine datasets covering a wide age range and including individuals with brain tumors. In comparison to AI-CT-derived labels from 37 MRI-CT pairs, TissUnet achieved a median Dice coefficient of 0.79 [IQR: 0.77-0.81] in a healthy adult cohort. In a second validation using expert manual annotations, median Dice was 0.83 [IQR: 0.83-0.84] in healthy individuals and 0.81 [IQR: 0.78-0.83] in tumor cases, outperforming previous state-of-the-art method. Acceptability testing resulted in an 89% acceptance rate after adjudication by a tie-breaker(N=108 MRIs), and TissUnet demonstrated excellent performance in the blinded comparative review (N=45 MRIs), including both healthy and tumor cases in pediatric populations. TissUnet enables fast, accurate, and reproducible segmentation of extracranial tissues, supporting large-scale studies on craniofacial morphology, treatment effects, and cardiometabolic risk using standard brain T1w MRI.
Authors: Yuhan Hao, Zhengning Li, Lei Sun, Weilong Wang, Naixin Yi, Sheng Song, Caihong Qin, Mofan Zhou, Yifei Zhan, Peng Jia, Xianpeng Lang
Abstract: Vision-Language-Action (VLA) models have advanced autonomous driving, but existing benchmarks still lack scenario diversity, reliable action-level annotation, and evaluation protocols aligned with human preferences. To address these limitations, we introduce DriveAction, the first action-driven benchmark specifically designed for VLA models, comprising 16,185 QA pairs generated from 2,610 driving scenarios. DriveAction leverages real-world driving data proactively collected by users of production-level autonomous vehicles to ensure broad and representative scenario coverage, offers high-level discrete action labels collected directly from users' actual driving operations, and implements an action-rooted tree-structured evaluation framework that explicitly links vision, language, and action tasks, supporting both comprehensive and task-specific assessment. Our experiments demonstrate that state-of-the-art vision-language models (VLMs) require both vision and language guidance for accurate action prediction: on average, accuracy drops by 3.3% without vision input, by 4.1% without language input, and by 8.0% without either. Our evaluation supports precise identification of model bottlenecks with robust and consistent results, thus providing new insights and a rigorous foundation for advancing human-like decisions in autonomous driving.
Authors: Sajjad Abdoli, Freeman Lewin, Gediminas Vasiliauskas, Fabian Schonholz
Abstract: The development of modern Artificial Intelligence (AI) models, particularly diffusion-based models employed in computer vision and image generation tasks, is undergoing a paradigmatic shift in development methodologies. Traditionally dominated by a "Model Centric" approach, in which performance gains were primarily pursued through increasingly complex model architectures and hyperparameter optimization, the field is now recognizing a more nuanced "Data-Centric" approach. This emergent framework foregrounds the quality, structure, and relevance of training data as the principal driver of model performance. To operationalize this paradigm shift, we introduce the DataSeeds.AI sample dataset (the "DSD"), initially comprised of approximately 10,610 high-quality human peer-ranked photography images accompanied by extensive multi-tier annotations. The DSD is a foundational computer vision dataset designed to usher in a new standard for commercial image datasets. Representing a small fraction of DataSeed.AI's 100 million-plus image catalog, the DSD provides a scalable foundation necessary for robust commercial and multimodal AI development. Through this in-depth exploratory analysis, we document the quantitative improvements generated by the DSD on specific models against known benchmarks and make the code and the trained models used in our evaluation publicly available.
Authors: Tailin Zhou, Zhilin Chen, Wenlong Lyu, Zhitang Chen, Danny H. K. Tsang, Jun Zhang
Abstract: Optimizing complex systems, from discovering therapeutic drugs to designing high-performance materials, remains a fundamental challenge across science and engineering, as the underlying rules are often unknown and costly to evaluate. Offline optimization aims to optimize designs for target scores using pre-collected datasets without system interaction. However, conventional approaches may fail beyond training data, predicting inaccurate scores and generating inferior designs. This paper introduces ManGO, a diffusion-based framework that learns the design-score manifold, capturing the design-score interdependencies holistically. Unlike existing methods that treat design and score spaces in isolation, ManGO unifies forward prediction and backward generation, attaining generalization beyond training data. Key to this is its derivative-free guidance for conditional generation, coupled with adaptive inference-time scaling that dynamically optimizes denoising paths. Extensive evaluations demonstrate that ManGO outperforms 24 single- and 10 multi-objective optimization methods across diverse domains, including synthetic tasks, robot control, material design, DNA sequence, and real-world engineering optimization.
Authors: Fardis Nadimi, Payam Abdisarabshali, Kasra Borazjani, Jacob Chakareski, Seyyedali Hosseinalipour
Abstract: Extended reality (XR) systems, which consist of virtual reality (VR), augmented reality (AR), and mixed reality (XR), offer a transformative interface for immersive, multi-modal, and embodied human-computer interaction. In this paper, we envision that multi-modal multi-task (M3T) federated foundation models (FedFMs) can offer transformative capabilities for XR systems through integrating the representational strength of M3T foundation models (FMs) with the privacy-preserving model training principles of federated learning (FL). We present a modular architecture for FedFMs, which entails different coordination paradigms for model training and aggregations. Central to our vision is the codification of XR challenges that affect the implementation of FedFMs under the SHIFT dimensions: (1) Sensor and modality diversity, (2) Hardware heterogeneity and system-level constraints, (3) Interactivity and embodied personalization, (4) Functional/task variability, and (5) Temporality and environmental variability. We illustrate the manifestation of these dimensions across a set of emerging and anticipated applications of XR systems. Finally, we propose evaluation metrics, dataset requirements, and design tradeoffs necessary for the development of resource-aware FedFMs in XR. This perspective aims to chart the technical and conceptual foundations for context-aware privacy-preserving intelligence in the next generation of XR systems.
Authors: Xinghang Li, Jingzhe Ding, Chao Peng, Bing Zhao, Xiang Gao, Hongwan Gao, Xinchen Gu
Abstract: The code generation capabilities of large language models(LLMs) have emerged as a critical dimension in evaluating their overall performance. However, prior research has largely overlooked the security risks inherent in the generated code. In this work, we introduce \benchmark, a benchmark specifically designed to assess the security of LLM-generated code. The dataset encompasses a wide range of common software development scenarios and vulnerability types. Building upon this benchmark, we develop an automatic evaluation framework that leverages both static application security testing(SAST) and LLM-based judging to assess the presence of security vulnerabilities in model-generated code. Through the empirical evaluation of state-of-the-art LLMs on \benchmark, we reveal notable deficiencies in their ability to produce vulnerability-free code. Our findings highlight pressing challenges and offer actionable insights for future advancements in the secure code generation performance of LLMs. The data and code will be released soon.
Authors: Ramteja Sajja, Yusuf Sermet, Brian Fodale, Ibrahim Demir
Abstract: As generative AI tools become increasingly integrated into higher education, understanding how students interact with and perceive these technologies is essential for responsible and effective adoption. This study evaluates the use of the Educational AI Hub, an AI-powered learning framework, in undergraduate civil and environmental engineering courses at a large R1 public university. Using a mixed-methods approach that combines pre- and post-surveys, system usage logs, and qualitative analysis of the open-ended prompts and questions students posed to the AI chatbot, the research explores students' perceptions of trust, ethical concerns, usability, and learning outcomes. Findings reveal that students appreciated the AI assistant for its convenience and comfort, with nearly half reporting greater ease in using the AI tool compared to seeking help from instructors or teaching assistants. The tool was seen as most helpful for completing homework and understanding course concepts, though perceptions of its instructional quality were mixed. Ethical concerns emerged as a key barrier to full engagement: while most students viewed AI use as ethically acceptable, many expressed uncertainties about institutional policies and apprehension about potential academic misconduct. This study contributes to the growing body of research on AI in education by highlighting the importance of usability, policy clarity, and faculty guidance in fostering meaningful AI engagement. The findings suggest that while students are ready to embrace AI as a supplement to human instruction, thoughtful integration and transparent institutional frameworks are critical for ensuring student confidence, trust, and learning effectiveness.
Authors: Yan Wang, Yueru He, Ruoyu Xiang, Jeff Zhao
Abstract: Recent advances in large language models (LLMs) hold great promise for financial applications but introduce critical accuracy and compliance challenges in Digital Regulatory Reporting (DRR). To address these issues, we propose RKEFino1, a regulation knowledge-enhanced financial reasoning model built upon Fino1, fine-tuned with domain knowledge from XBRL, CDM, and MOF. We formulate two QA tasks-knowledge-based and mathematical reasoning-and introduce a novel Numerical NER task covering financial entities in both sentences and tables. Experimental results demonstrate the effectiveness and generalization capacity of RKEFino1 in compliance-critical financial tasks. We have released our model on Hugging Face.
Authors: Chaofan Pan, Jiafen Liu, Yanhua Li, Linbo Xiong, Fan Min, Wei Wei, Xin Yang
Abstract: Continual Learning (CL) is a powerful tool that enables agents to learn a sequence of tasks, accumulating knowledge learned in the past and using it for problem-solving or future task learning. However, existing CL methods often assume that the agent's capabilities remain static within dynamic environments, which doesn't reflect real-world scenarios where capabilities dynamically change. This paper introduces a new and realistic problem: Continual Learning with Dynamic Capabilities (CL-DC), posing a significant challenge for CL agents: How can policy generalization across different action spaces be achieved? Inspired by the cortical functions, we propose an Action-Adaptive Continual Learning framework (AACL) to address this challenge. Our framework decouples the agent's policy from the specific action space by building an action representation space. For a new action space, the encoder-decoder of action representations is adaptively fine-tuned to maintain a balance between stability and plasticity. Furthermore, we release a benchmark based on three environments to validate the effectiveness of methods for CL-DC. Experimental results demonstrate that our framework outperforms popular methods by generalizing the policy across action spaces.
Authors: Adrian Ly, Richard Dazeley, Peter Vamplew, Francisco Cruz, Sunil Aryal
Abstract: While many algorithmic extensions to Deep Q-Networks (DQN) have been proposed, there remains limited understanding of how different improvements interact. In particular, multi-step and ensemble style extensions have shown promise in reducing overestimation bias, thereby improving sample efficiency and algorithmic stability. In this paper, we introduce a novel algorithm called Ensemble Elastic Step DQN (EEDQN), which unifies ensembles with elastic step updates to stabilise algorithmic performance. EEDQN is designed to address two major challenges in deep reinforcement learning: overestimation bias and sample efficiency. We evaluated EEDQN against standard and ensemble DQN variants across the MinAtar benchmark, a set of environments that emphasise behavioral learning while reducing representational complexity. Our results show that EEDQN achieves consistently robust performance across all tested environments, outperforming baseline DQN methods and matching or exceeding state-of-the-art ensemble DQNs in final returns on most of the MinAtar environments. These findings highlight the potential of systematically combining algorithmic improvements and provide evidence that ensemble and multi-step methods, when carefully integrated, can yield substantial gains.
Authors: Pascal Jr Tikeng Notsawo, Guillaume Dumas, Guillaume Rabusseau
Abstract: Grokking refers to a delayed generalization following overfitting when optimizing artificial neural networks with gradient-based methods. In this work, we demonstrate that grokking can be induced by regularization, either explicit or implicit. More precisely, we show that when there exists a model with a property $P$ (e.g., sparse or low-rank weights) that generalizes on the problem of interest, gradient descent with a small but non-zero regularization of $P$ (e.g., $\ell_1$ or nuclear norm regularization) results in grokking. This extends previous work showing that small non-zero weight decay induces grokking. Moreover, our analysis shows that over-parameterization by adding depth makes it possible to grok or ungrok without explicitly using regularization, which is impossible in shallow cases. We further show that the $\ell_2$ norm is not a reliable proxy for generalization when the model is regularized toward a different property $P$, as the $\ell_2$ norm grows in many cases where no weight decay is used, but the model generalizes anyway. We also show that grokking can be amplified solely through data selection, with any other hyperparameter fixed.
Authors: Dumindu Tissera, Omar Awadallah, Muhammad Umair Danish, Ayan Sadhu, Katarina Grolinger
Abstract: Multi-label Classification (MLC) assigns an instance to one or more non-exclusive classes. A challenge arises when the dataset contains a large proportion of instances with no assigned class, referred to as negative data, which can overwhelm the learning process and hinder the accurate identification and classification of positive instances. Nevertheless, it is common in MLC applications such as industrial defect detection, agricultural disease identification, and healthcare diagnosis to encounter large amounts of negative data. Assigning a separate negative class to these instances further complicates the learning objective and introduces unnecessary redundancies. To address this challenge, we redesign standard MLC loss functions by deriving a likelihood of any class being present, formulated by a normalized weighted geometric mean of the predicted class probabilities. We introduce a regularization parameter that controls the relative contribution of the absent class probabilities to the any-class presence likelihood in positive instances. The any-class presence likelihood complements the multi-label learning by encouraging the network to become more aware of implicit positive instances and improve the label classification within those positive instances. Experiments on large-scale datasets with negative data: SewerML, modified COCO, and ChestX-ray14, across various networks and base loss functions show that our loss functions consistently improve MLC performance of their standard loss counterparts, achieving gains of up to 6.01 percentage points in F1, 8.06 in F2, and 3.11 in mean average precision, all without additional parameters or computational complexity. Code available at: https://github.com/ML-for-Sensor-Data-Western/gmean-mlc
URLs: https://github.com/ML-for-Sensor-Data-Western/gmean-mlc
Authors: Fang Wu, Vijay Prakash Dwivedi, Jure Leskovec
Abstract: Large language models (LLMs) have demonstrated remarkable capabilities across various domains, yet their application to relational deep learning (RDL) remains underexplored. Existing approaches adapt LLMs by traversing relational links between entities in a database and converting the structured data into flat text documents. Still, this text-based serialization disregards critical relational structures, introduces redundancy, and often exceeds standard LLM context lengths. We introduce Rel-LLM, a novel architecture that utilizes a graph neural network (GNN)- based encoder to generate structured relational prompts for LLMs within a retrieval-augmented generation (RAG) framework. Unlike traditional text-based serialization approaches, our method preserves the inherent relational structure of databases while enabling LLMs to effectively process and reason over complex entity relationships. Specifically, the GNN encoder extracts a local subgraph around an entity to build feature representations that contain relevant entity relationships and temporal dependencies. These representations are transformed into structured prompts using a denormalization process, effectively allowing the LLM to reason over relational structures. Through extensive experiments, we demonstrate that Rel-LLM outperforms existing methods on key RDL tasks, offering a scalable and efficient approach to integrating LLMs with structured data sources. Code is available at https://github.com/smiles724/Rel-LLM.
Authors: En Yu, Jie Lu, Guangquan Zhang
Abstract: Real-world data streams exhibit inherent non-stationarity characterized by concept drift, posing significant challenges for adaptive learning systems. While existing methods address isolated distribution shifts, they overlook the critical co-evolution of label spaces and distributions under limited supervision and persistent uncertainty. To address this, we formalize Generalized Incremental Learning under Concept Drift (GILCD), characterizing the joint evolution of distributions and label spaces in open-environment streaming contexts, and propose a novel framework called Calibrated Source-Free Adaptation (CSFA). First, CSFA introduces a training-free prototype calibration mechanism that dynamically fuses emerging prototypes with base representations, enabling stable new-class identification without optimization overhead. Second, we design a novel source-free adaptation algorithm, i.e., Reliable Surrogate Gap Sharpness-aware (RSGS) minimization. It integrates sharpness-aware perturbation loss optimization with surrogate gap minimization, while employing entropy-based uncertainty filtering to discard unreliable samples. This mechanism ensures robust distribution alignment and mitigates generalization degradation caused by uncertainties. Therefore, CSFA establishes a unified framework for stable adaptation to evolving semantics and distributions in open-world streaming scenarios. Extensive experiments validate the superior performance and effectiveness of CSFA compared to state-of-the-art approaches.
Authors: Zhilong Wang, Neha Nagaraja, Lan Zhang, Hayretdin Bahsi, Pawan Patil, Peng Liu
Abstract: LLM agents are widely used as agents for customer support, content generation, and code assistance. However, they are vulnerable to prompt injection attacks, where adversarial inputs manipulate the model's behavior. Traditional defenses like input sanitization, guard models, and guardrails are either cumbersome or ineffective. In this paper, we propose a novel, lightweight defense mechanism called Polymorphic Prompt Assembling (PPA), which protects against prompt injection with near-zero overhead. The approach is based on the insight that prompt injection requires guessing and breaking the structure of the system prompt. By dynamically varying the structure of system prompts, PPA prevents attackers from predicting the prompt structure, thereby enhancing security without compromising performance. We conducted experiments to evaluate the effectiveness of PPA against existing attacks and compared it with other defense methods.
Authors: Ruining Sun, Hongsheng Hu, Wei Luo, Zhaoxi Zhang, Yanjun Zhang, Haizhuan Yuan, Leo Yu Zhang
Abstract: With the rapid advancement of deep learning technology, pre-trained encoder models have demonstrated exceptional feature extraction capabilities, playing a pivotal role in the research and application of deep learning. However, their widespread use has raised significant concerns about the risk of training data privacy leakage. This paper systematically investigates the privacy threats posed by membership inference attacks (MIAs) targeting encoder models, focusing on contrastive learning frameworks. Through experimental analysis, we reveal the significant impact of model architecture complexity on membership privacy leakage: As more advanced encoder frameworks improve feature-extraction performance, they simultaneously exacerbate privacy-leakage risks. Furthermore, this paper proposes a novel membership inference attack method based on the p-norm of feature vectors, termed the Embedding Lp-Norm Likelihood Attack (LpLA). This method infers membership status, by leveraging the statistical distribution characteristics of the p-norm of feature vectors. Experimental results across multiple datasets and model architectures demonstrate that LpLA outperforms existing methods in attack performance and robustness, particularly under limited attack knowledge and query volumes. This study not only uncovers the potential risks of privacy leakage in contrastive learning frameworks, but also provides a practical basis for privacy protection research in encoder models. We hope that this work will draw greater attention to the privacy risks associated with self-supervised learning models and shed light on the importance of a balance between model utility and training data privacy. Our code is publicly available at: https://github.com/SeroneySun/LpLA_code.
Authors: Rudransh Agnihotri, Ananya Pandey
Abstract: Reward-model training is the cost bottleneck in modern Reinforcement Learning Human Feedback (RLHF) pipelines, often requiring tens of billions of parameters and an offline preference-tuning phase. In the proposed method, a frozen, instruction-tuned 7B LLM is augmented with only a one line JSON rubric and a rank-16 LoRA adapter (affecting just 0.8% of the model's parameters), enabling it to serve as a complete substitute for the previously used heavyweight evaluation models. The plug-and-play judge achieves 96.2% accuracy on RewardBench, outperforming specialized reward networks ranging from 27B to 70B parameters. Additionally, it allows a 7B actor to outperform the top 70B DPO baseline, which scores 61.8%, by achieving 92% exact match accuracy on GSM-8K utilizing online PPO. Thorough ablations indicate that (i) six in context demonstrations deliver the majority of the zero-to-few-shot improvements (+2pp), and (ii) the LoRA effectively addresses the remaining disparity, particularly in the safety and adversarial Chat-Hard segments. The proposed model introduces HH-Rationales, a subset of 10,000 pairs from Anthropic HH-RLHF, to examine interpretability, accompanied by human generated justifications. GPT-4 scoring indicates that our LoRA judge attains approximately = 9/10 in similarity to human explanations, while zero-shot judges score around =5/10. These results indicate that the combination of prompt engineering and tiny LoRA produces a cost effective, transparent, and easily adjustable reward function, removing the offline phase while achieving new state-of-the-art outcomes for both static evaluation and online RLHF.
Authors: Antrea Christou, Chris Davis Jaldi, Joseph Zalewski, Hande K\"u\c{c}\"uk McGinty, Pascal Hitzler, Cogan Shimizu
Abstract: Educational, learning, and training materials have become extremely commonplace across the Internet. Yet, they frequently remain disconnected from each other, fall into platform silos, and so on. One way to overcome this is to provide a mechanism to integrate the material and provide cross-links across topics. In this paper, we present the Curriculum KG Ontology, which we use as a framework for the dense interlinking of educational materials, by first starting with organizational and broad pedagogical principles. We provide a materialized graph for the Prototype Open Knowledge Network use-case, and validate it using competency questions sourced from domain experts and educators.
Authors: Zhongying Wang, Thoai D. Ngo, Hamidreza Zoraghein, Benjamin Lucas, Morteza Karimzadeh
Abstract: The COVID-19 pandemic's severe impact highlighted the need for accurate, timely hospitalization forecasting to support effective healthcare planning. However, most forecasting models struggled, especially during variant surges, when they were needed most. This study introduces a novel Long Short-Term Memory (LSTM) framework for forecasting daily state-level incident hospitalizations in the United States. We present a spatiotemporal feature, Social Proximity to Hospitalizations (SPH), derived from Facebook's Social Connectedness Index to improve forecasts. SPH serves as a proxy for interstate population interaction, capturing transmission dynamics across space and time. Our parallel LSTM architecture captures both short- and long-term temporal dependencies, and our multi-horizon ensembling strategy balances consistency and forecasting error. Evaluation against COVID-19 Forecast Hub ensemble models during the Delta and Omicron surges reveals superiority of our model. On average, our model surpasses the ensemble by 27, 42, 54, and 69 hospitalizations per state on the $7^{th}$, $14^{th}$, $21^{st}$, and $28^{th}$ forecast days, respectively, during the Omicron surge. Data-ablation experiments confirm SPH's predictive power, highlighting its effectiveness in enhancing forecasting models. This research not only advances hospitalization forecasting but also underscores the significance of spatiotemporal features, such as SPH, in refining predictive performance in modeling the complex dynamics of infectious disease spread.
Authors: Yang Li, Zhi Chen
Abstract: Optimal execution in financial markets refers to the process of strategically transacting a large volume of assets over a period to achieve the best possible outcome by balancing the trade-off between market impact costs and timing or volatility risks. Traditional optimal execution strategies, such as static Almgren-Chriss models, often prove suboptimal in dynamic financial markets. This paper propose flowOE, a novel imitation learning framework based on flow matching models, to address these limitations. FlowOE learns from a diverse set of expert traditional strategies and adaptively selects the most suitable expert behavior for prevailing market conditions. A key innovation is the incorporation of a refining loss function during the imitation process, enabling flowOE not only to mimic but also to improve upon the learned expert actions. To the best of our knowledge, this work is the first to apply flow matching models in a stochastic optimal execution problem. Empirical evaluations across various market conditions demonstrate that flowOE significantly outperforms both the specifically calibrated expert models and other traditional benchmarks, achieving higher profits with reduced risk. These results underscore the practical applicability and potential of flowOE to enhance adaptive optimal execution.
Authors: Guang-Xing Li
Abstract: Physics has been transforming our view of nature for centuries. While combining physical knowledge with computational approaches has enabled detailed modeling of physical systems' evolution, understanding the emergence of patterns and structures remains limited. Correlations between quantities are the most reliable approach to describe relationships between different variables. However, for complex patterns, directly searching for correlations is often impractical, as complexity and spatial inhomogeneity can obscure correlations. We discovered that the key is to search for correlations in local regions and developed a new method, adjacent correlation analysis, to extract such correlations and represent them in phase space. When multiple observations are available, a useful way to study a system is to analyze distributions in phase space using the Probability Density Function (PDF). Adjacent correlation analysis evaluates vectors representing local correlations, which can be overlaid on the PDF plot to form the adjacent correlation plot. These correlation vectors often exhibit remarkably regular patterns and may lead to the discovery of new laws. The vectors we derive are equivalent to the vector field in dynamical systems on the attracting manifold. By efficiently representing spatial patterns as correlation vectors in phase space, our approach opens avenues for classification, prediction, parameter fitting, and forecasting.
Authors: Bi Huo, Bin Tu, Cheng Qin, Da Zheng, Debing Zhang, Dongjie Zhang, En Li, Fu Guo, Jian Yao, Jie Lou, Junfeng Tian, Li Hu, Ran Zhu, Shengdong Chen, Shuo Liu, Su Guang, Te Wo, Weijun Zhang, Xiaoming Shi, Xinxin Peng, Xing Wu, Yawen Liu, Yuqiu Ji, Ze Wen, Zhenhai Liu, Zichao Li, Zilong Liao
Abstract: Mixture of Experts (MoE) models have emerged as a promising paradigm for scaling language models efficiently by activating only a subset of parameters for each input token. In this report, we present dots.llm1, a large-scale MoE model that activates 14B parameters out of a total of 142B parameters, delivering performance on par with state-of-the-art models while reducing training and inference costs. Leveraging our meticulously crafted and efficient data processing pipeline, dots.llm1 achieves performance comparable to Qwen2.5-72B after pretraining on 11.2T high-quality tokens and post-training to fully unlock its capabilities. Notably, no synthetic data is used during pretraining. To foster further research, we open-source intermediate training checkpoints at every one trillion tokens, providing valuable insights into the learning dynamics of large language models.
Authors: Meng Fan, Yifan Zuo, Patrick Blaes, Harley Montgomery, Subhasis Das
Abstract: Sensor fusion is crucial for a performant and robust Perception system in autonomous vehicles, but sensor staleness, where data from different sensors arrives with varying delays, poses significant challenges. Temporal misalignment between sensor modalities leads to inconsistent object state estimates, severely degrading the quality of trajectory predictions that are critical for safety. We present a novel and model-agnostic approach to address this problem via (1) a per-point timestamp offset feature (for LiDAR and radar both relative to camera) that enables fine-grained temporal awareness in sensor fusion, and (2) a data augmentation strategy that simulates realistic sensor staleness patterns observed in deployed vehicles. Our method is integrated into a perspective-view detection model that consumes sensor data from multiple LiDARs, radars and cameras. We demonstrate that while a conventional model shows significant regressions when one sensor modality is stale, our approach reaches consistently good performance across both synchronized and stale conditions.
Authors: Yogesh Verma, Amauri H. Souza, Vikas Garg
Abstract: The local inductive bias of message-passing graph neural networks (GNNs) hampers their ability to exploit key structural information (e.g., connectivity and cycles). Positional encoding (PE) and Persistent Homology (PH) have emerged as two promising approaches to mitigate this issue. PE schemes endow GNNs with location-aware features, while PH methods enhance GNNs with multiresolution topological features. However, a rigorous theoretical characterization of the relative merits and shortcomings of PE and PH has remained elusive. We bridge this gap by establishing that neither paradigm is more expressive than the other, providing novel constructions where one approach fails but the other succeeds. Our insights inform the design of a novel learnable method, PiPE (Persistence-informed Positional Encoding), which is provably more expressive than both PH and PE. PiPE demonstrates strong performance across a variety of tasks (e.g., molecule property prediction, graph classification, and out-of-distribution generalization), thereby advancing the frontiers of graph representation learning. Code is available at https://github.com/Aalto-QuML/PIPE.
Authors: Quansong He, Xiangde Min, Kaishen Wang, Tao He
Abstract: Medical image segmentation is a critical task in computer vision, with UNet serving as a milestone architecture. The typical component of UNet family is the skip connection, however, their skip connections face two significant limitations: (1) they lack effective interaction between features at different scales, and (2) they rely on simple concatenation or addition operations, which constrain efficient information integration. While recent improvements to UNet have focused on enhancing encoder and decoder capabilities, these limitations remain overlooked. To overcome these challenges, we propose a novel multi-scale feature fusion method that reimagines the UNet decoding process as solving an initial value problem (IVP), treating skip connections as discrete nodes. By leveraging principles from the linear multistep method, we propose an adaptive ordinary differential equation method to enable effective multi-scale feature fusion. Our approach is independent of the encoder and decoder architectures, making it adaptable to various U-Net-like networks. Experiments on ACDC, KiTS2023, MSD brain tumor, and ISIC2017/2018 skin lesion segmentation datasets demonstrate improved feature utilization, reduced network parameters, and maintained high performance. The code is available at https://github.com/nayutayuki/FuseUNet.
Authors: Yihan Xie, Sijing Li, Tianwei Lin, Zhuonan Wang, Chenglin Yang, Yu Zhong, Wenqiao Zhang, Haoyuan Li, Hao Jiang, Fengda Zhang, Qishan Chen, Jun Xiao, Yueting Zhuang, Beng Chin Ooi
Abstract: We present Heartcare Suite, a multimodal comprehensive framework for finegrained electrocardiogram (ECG) understanding. It comprises three key components: (i) Heartcare-220K, a high-quality, structured, and comprehensive multimodal ECG dataset covering essential tasks such as disease diagnosis, waveform morphology analysis, and rhythm interpretation. (ii) Heartcare-Bench, a systematic and multi-dimensional benchmark designed to evaluate diagnostic intelligence and guide the optimization of Medical Multimodal Large Language Models (Med-MLLMs) in ECG scenarios. and (iii) HeartcareGPT with a tailored tokenizer Bidirectional ECG Abstract Tokenization (Beat), which compresses raw multi-lead signals into semantically rich discrete tokens via duallevel vector quantization and query-guided bidirectional diffusion mechanism. Built upon Heartcare-220K, HeartcareGPT achieves strong generalization and SoTA performance across multiple clinically meaningful tasks. Extensive experiments demonstrate that Heartcare Suite is highly effective in advancing ECGspecific multimodal understanding and evaluation. Our project is available at https://github.com/Wznnnnn/Heartcare-Suite .
Authors: Yiwen Ding, Krishna Manoorkar
Abstract: Recently, description logic LE-ALC was introduced for reasoning in the semantic environment of enriched formal contexts, and a polynomial-time tableaux algorithm was developed to check the consistency of knowledge bases with acyclic TBoxes. In this work, we introduce a fuzzy generalization of LE-ALC called LE-FALC which provides a description logic counterpart of many-valued normal non-distributive logic a.k.a. many-valued LE-logic. This description logic can be used to represent and reason about knowledge in the formal framework of fuzzy formal contexts and fuzzy formal concepts. We provide a tableaux algorithm that provides a complete and sound polynomial-time decision procedure to check the consistency of LE-FALC ABoxes. As a result, we also obtain an exponential-time decision procedure for checking the consistency of LE-FALC with acyclic TBoxes by unraveling.
Authors: Sandro Preto (Federal University of ABC, Brazil), Marcelo Finger (University of Sao Paulo, Brazil)
Abstract: A possible path to the interpretability of neural networks is to (approximately) represent them in the regional format of piecewise linear functions, where regions of inputs are associated to linear functions computing the network outputs. We present an algorithm for the translation of feedforward neural networks with ReLU activation functions in hidden layers and truncated identity activation functions in the output layer. We also empirically investigate the complexity of regional representations outputted by our method for neural networks with varying sizes. Lattice and logical representations of neural networks are straightforward from regional representations as long as they satisfy a specific property. So we empirically investigate to what extent the translations by our algorithm satisfy such property.
Authors: Cheonbok Park, Jeonghoon Kim, Joosung Lee, Sanghwan Bae, Jaegul Choo, Kangmin Yoo
Abstract: We identify \textbf{Cross-lingual Collapse}, a systematic drift in which the chain-of-thought (CoT) of a multilingual language model reverts to its dominant pre-training language even when the prompt is expressed in a different language. Recent large language models (LLMs) with reinforcement learning with verifiable reward (RLVR) have achieved strong logical reasoning performances by exposing their intermediate reasoning traces, giving rise to large reasoning models (LRMs). However, the mechanism behind multilingual reasoning in LRMs is not yet fully explored. To investigate the issue, we fine-tune multilingual LRMs with Group-Relative Policy Optimization (GRPO) on translated versions of the GSM$8$K and SimpleRL-Zoo datasets in three different languages: Chinese, Korean, and Ukrainian. During training, we monitor both task accuracy and language consistency of the reasoning chains. Our experiments reveal three key findings: (i) GRPO rapidly amplifies pre-training language imbalances, leading to the erosion of low-resource languages within just a few hundred updates; (ii) language consistency reward mitigates this drift but does so at the expense of an almost 5 - 10 pp drop in accuracy. and (iii) the resulting language collapse is severely damaging and largely irreversible, as subsequent fine-tuning struggles to steer the model back toward its original target-language reasoning capabilities. Together, these findings point to a remarkable conclusion: \textit{not all languages are trained equally for reasoning}. Furthermore, our paper sheds light on the roles of reward shaping, data difficulty, and pre-training priors in eliciting multilingual reasoning.
Authors: Marcel Klemt, Carlotta Segna, Anna Rohrbach
Abstract: Generative AI advances rapidly, allowing the creation of very realistic manipulated video and audio. This progress presents a significant security and ethical threat, as malicious users can exploit DeepFake techniques to spread misinformation. Recent DeepFake detection approaches explore the multimodal (audio-video) threat scenario. In particular, there is a lack of reproducibility and critical issues with existing datasets - such as the recently uncovered silence shortcut in the widely used FakeAVCeleb dataset. Considering the importance of this topic, we aim to gain a deeper understanding of the key issues affecting benchmarking in audio-video DeepFake detection. We examine these challenges through the lens of the three core benchmarking pillars: datasets, detection methods, and evaluation protocols. To address these issues, we spotlight the recent DeepSpeak v1 dataset and are the first to propose an evaluation protocol and benchmark it using SOTA models. We introduce SImple Multimodal BAseline (SIMBA), a competitive yet minimalistic approach that enables the exploration of diverse design choices. We also deepen insights into the issue of audio shortcuts and present a promising mitigation strategy. Finally, we analyze and enhance the evaluation scheme on the widely used FakeAVCeleb dataset. Our findings offer a way forward in the complex area of audio-video DeepFake detection.
Authors: Yuqian Fu, Runze Wang, Yanwei Fu, Danda Pani Paudel, Luc Van Gool
Abstract: In this report, we present a cross-view multi-modal object segmentation approach for the object correspondence task in the Ego-Exo4D Correspondence Challenges 2025. Given object queries from one perspective (e.g., ego view), the goal is to predict the corresponding object masks in another perspective (e.g., exo view). To tackle this task, we propose a multimodal condition fusion module that enhances object localization by leveraging both visual masks and textual descriptions as segmentation conditions. Furthermore, to address the visual domain gap between ego and exo views, we introduce a cross-view object alignment module that enforces object-level consistency across perspectives, thereby improving the model's robustness to viewpoint changes. Our proposed method ranked second on the leaderboard of the large-scale Ego-Exo4D object correspondence benchmark. Code will be made available at https://github.com/lovelyqian/ObjectRelator.
Authors: Han Ji, Yuqi Feng, Jiahao Fan, Yanan Sun
Abstract: Evaluation is a critical but costly procedure in neural architecture search (NAS). Performance predictors have been widely adopted to reduce evaluation costs by directly estimating architecture performance. The effectiveness of predictors is heavily influenced by the choice of loss functions. While traditional predictors employ regression loss functions to evaluate the absolute accuracy of architectures, recent approaches have explored various ranking-based loss functions, such as pairwise and listwise ranking losses, to focus on the ranking of architecture performance. Despite their success in NAS, the effectiveness and characteristics of these loss functions have not been thoroughly investigated. In this paper, we conduct the first comprehensive study on loss functions in performance predictors, categorizing them into three main types: regression, ranking, and weighted loss functions. Specifically, we assess eight loss functions using a range of NAS-relevant metrics on 13 tasks across five search spaces. Our results reveal that specific categories of loss functions can be effectively combined to enhance predictor-based NAS. Furthermore, our findings could provide practical guidance for selecting appropriate loss functions for various tasks. We hope this work provides meaningful insights to guide the development of loss functions for predictor-based methods in the NAS community.
Authors: Yushang Zhao, Yike Peng, Dannier Li, Yuxin Yang, Chengrui Zhou, Jing Dong
Abstract: With the rapid growth of fintech, personalized financial product recommendations have become increasingly important. Traditional methods like collaborative filtering or content-based models often fail to capture users' latent preferences and complex relationships. We propose a hybrid framework integrating large language models (LLMs) and graph neural networks (GNNs). A pre-trained LLM encodes text data (e.g., user reviews) into rich feature vectors, while a heterogeneous user-product graph models interactions and social ties. Through a tailored message-passing mechanism, text and graph information are fused within the GNN to jointly optimize embeddings. Experiments on public and real-world financial datasets show our model outperforms standalone LLM or GNN in accuracy, recall, and NDCG, with strong interpretability. This work offers new insights for personalized financial recommendations and cross-modal fusion in broader recommendation tasks.
Authors: Yue Lin, Shuhui Zhu, William A Cunningham, Wenhao Li, Pascal Poupart, Hongyuan Zha, Baoxiang Wang
Abstract: Bayesian persuasion, an extension of cheap-talk communication, involves an informed sender committing to a signaling scheme to influence a receiver's actions. Compared to cheap talk, this sender's commitment enables the receiver to verify the incentive compatibility of signals beforehand, facilitating cooperation. While effective in one-shot scenarios, Bayesian persuasion faces computational complexity (NP-hardness) when extended to long-term interactions, where the receiver may adopt dynamic strategies conditional on past outcomes and future expectations. To address this complexity, we introduce the bargaining perspective, which allows: (1) a unified framework and well-structured solution concept for long-term persuasion, with desirable properties such as fairness and Pareto efficiency; (2) a clear distinction between two previously conflated advantages: the sender's informational advantage and first-proposer advantage. With only modest modifications to the standard setting, this perspective makes explicit the common knowledge of the game structure and grants the receiver comparable commitment capabilities, thereby reinterpreting classic one-sided persuasion as a balanced information bargaining framework. The framework is validated through a two-stage validation-and-inference paradigm: We first demonstrate that GPT-o3 and DeepSeek-R1, out of publicly available LLMs, reliably handle standard tasks; We then apply them to persuasion scenarios to test that the outcomes align with what our information-bargaining framework suggests. All code, results, and terminal logs are publicly available at github.com/YueLin301/InformationBargaining.
Authors: Daming Wang, Yuhao Song, Zijian He, Kangliang Chen, Xing Pan, Lu Deng, Weihao Gu
Abstract: We present HaoMo Vision-Language Model (HMVLM), an end-to-end driving framework that implements the slow branch of a cognitively inspired fast-slow architecture. A fast controller outputs low-level steering, throttle, and brake commands, while a slow planner-a large vision-language model-generates high-level intents such as "yield to pedestrian" or "merge after the truck" without compromising latency. HMVLM introduces three upgrades: (1) selective five-view prompting with an embedded 4s history of ego kinematics, (2) multi-stage chain-of-thought (CoT) prompting that enforces a Scene Understanding -> Driving Decision -> Trajectory Inference reasoning flow, and (3) spline-based trajectory post-processing that removes late-stage jitter and sharp turns. Trained on the Waymo Open Dataset, these upgrades enable HMVLM to achieve a Rater Feedback Score (RFS) of 7.7367, securing 2nd place in the 2025 Waymo Vision-based End-to-End (E2E) Driving Challenge and surpassing the public baseline by 2.77%.
Authors: Chongshang Yan, Jiaxuan He, Delun Li, Yi Yang, Wenjie Song
Abstract: The zero-shot object navigation (ZSON) in unknown open-ended environments coupled with semantically novel target often suffers from the significant decline in performance due to the neglect of high-dimensional implicit scene information and the long-range target searching task. To address this, we proposed an active object navigation framework with Environmental Attributes Map (EAM) and MLLM Hierarchical Reasoning module (MHR) to improve its success rate and efficiency. EAM is constructed by reasoning observed environments with SBERT and predicting unobserved ones with Diffusion, utilizing human space regularities that underlie object-room correlations and area adjacencies. MHR is inspired by EAM to perform frontier exploration decision-making, avoiding the circuitous trajectories in long-range scenarios to improve path efficiency. Experimental results demonstrate that the EAM module achieves 64.5\% scene mapping accuracy on MP3D dataset, while the navigation task attains SPLs of 28.4\% and 26.3\% on HM3D and MP3D benchmarks respectively - representing absolute improvements of 21.4\% and 46.0\% over baseline methods.
Authors: Jakaria Islam Emon, Kazi Tamanna Alam, Md. Abu Salek
Abstract: Mean Opinion Score (MOS) prediction for text to music systems requires evaluating both overall musical quality and text prompt alignment. This paper introduces WhisQ, a multimodal architecture that addresses this dual-assessment challenge through sequence level co-attention and optimal transport regularization. WhisQ employs the Whisper Base pretrained model for temporal audio encoding and Qwen 3, a 0.6B Small Language Model (SLM), for text encoding, with both maintaining sequence structure for fine grained cross-modal modeling. The architecture features specialized prediction pathways: OMQ is predicted from pooled audio embeddings, while TA leverages bidirectional sequence co-attention between audio and text. Sinkhorn optimal transport loss further enforce semantic alignment in the shared embedding space. On the MusicEval Track-1 dataset, WhisQ achieves substantial improvements over the baseline: 7% improvement in Spearman correlation for OMQ and 14% for TA. Ablation studies reveal that optimal transport regularization provides the largest performance gain (10% SRCC improvement), demonstrating the importance of explicit cross-modal alignment for text-to-music evaluation.
Authors: Chenyang Shao, Xinyang Liu, Yutang Lin, Fengli Xu, Yong Li
Abstract: Multi-step reasoning has proven essential for enhancing the problem-solving capabilities of Large Language Models (LLMs) by decomposing complex tasks into intermediate steps, either explicitly or implicitly. Extending the reasoning chain at test time through deeper thought processes or broader exploration, can furthur improve performance, but often incurs substantial costs due to the explosion in token usage. Yet, many reasoning steps are relatively simple and can be handled by more efficient smaller-scale language models (SLMs). This motivates hybrid approaches that allocate subtasks across models of varying capacities. However, realizing such collaboration requires accurate task decomposition and difficulty-aware subtask allocation, which is challenging. To address this, we propose R2-Reasoner, a novel framework that enables collaborative reasoning across heterogeneous LLMs by dynamically routing sub-tasks based on estimated complexity. At the core of our framework is a Reinforced Model Router, composed of a task decomposer and a subtask allocator. The task decomposer segments complex input queries into logically ordered subtasks, while the subtask allocator assigns each subtask to the most appropriate model, ranging from lightweight SLMs to powerful LLMs, balancing accuracy and efficiency. To train this router, we introduce a staged pipeline that combines supervised fine-tuning on task-specific datasets with Group Relative Policy Optimization algorithm, enabling self-supervised refinement through iterative reinforcement learning. Extensive experiments across four challenging benchmarks demonstrate that R2-Reasoner reduces API costs by 86.85% while maintaining or surpassing baseline accuracy. Our framework paves the way for more cost-effective and adaptive LLM reasoning. The code is open-source at https://anonymous.4open.science/r/R2_Reasoner .
Authors: Steven Landgraf, Markus Hillemann, Markus Ulrich
Abstract: Semantic segmentation is critical for scene understanding but demands costly pixel-wise annotations, attracting increasing attention to semi-supervised approaches to leverage abundant unlabeled data. While semi-supervised segmentation is often promoted as a path toward scalable, real-world deployment, it is astonishing that current evaluation protocols exclusively focus on segmentation accuracy, entirely overlooking reliability and robustness. These qualities, which ensure consistent performance under diverse conditions (robustness) and well-calibrated model confidences as well as meaningful uncertainties (reliability), are essential for safety-critical applications like autonomous driving, where models must handle unpredictable environments and avoid sudden failures at all costs. To address this gap, we introduce the Reliable Segmentation Score (RSS), a novel metric that combines predictive accuracy, calibration, and uncertainty quality measures via a harmonic mean. RSS penalizes deficiencies in any of its components, providing an easy and intuitive way of holistically judging segmentation models. Comprehensive evaluations of UniMatchV2 against its predecessor and a supervised baseline show that semi-supervised methods often trade reliability for accuracy. While out-of-domain evaluations demonstrate UniMatchV2's robustness, they further expose persistent reliability shortcomings. We advocate for a shift in evaluation protocols toward more holistic metrics like RSS to better align semi-supervised learning research with real-world deployment needs.
Authors: Zarreen Reza, Alexander Mazur, Michael T. Dugdale, Robin Ray-Chaudhuri
Abstract: While Large Language Models (LLMs) are increasingly utilized as student-facing educational aids, their potential to directly support educators, particularly through locally deployable and customizable open-source solutions, remains significantly underexplored. Many existing educational solutions rely on cloud-based infrastructure or proprietary tools, which are costly and may raise privacy concerns. Regulated industries with limited budgets require affordable, self-hosted solutions. We introduce an end-to-end, open-source framework leveraging small (3B-7B parameters), locally deployed LLMs for customized teaching material generation and assessment. Our system uniquely incorporates an interactive loop crucial for effective small-model refinement, and an auxiliary LLM verifier to mitigate jailbreaking risks, enhancing output reliability and safety. Utilizing Retrieval and Context Augmented Generation (RAG/CAG), it produces factually accurate, customized pedagogically-styled content. Deployed on-premises for data privacy and validated through an evaluation pipeline and a college physics pilot, our findings show that carefully engineered small LLM systems can offer robust, affordable, practical, and safe educator support, achieving utility comparable to larger models for targeted tasks.
Authors: Bel\'en Ag\"uera-Marco, Itziar Gonzalez-Dios
Abstract: In this work, we present LengClaro2023, a dataset of legal-administrative texts in Spanish. Based on the most frequently used procedures from the Spanish Social Security website, we have created for each text two simplified equivalents. The first version follows the recommendations provided by arText claro. The second version incorporates additional recommendations from plain language guidelines to explore further potential improvements in the system. The linguistic resource created in this work can be used for evaluating automatic text simplification (ATS) systems in Spanish.
Authors: Jie Cao, Tianwei Lin, Hongyang He, Rolan Yan, Wenqiao Zhang, Juncheng Li, Dongping Zhang, Siliang Tang, Yueting Zhuang
Abstract: Recent studies integrate Low-Rank Adaptation (LoRA) and Mixture-of-Experts (MoE) to further enhance the performance of parameter-efficient fine-tuning (PEFT) methods in Large Language Model (LLM) applications. Existing methods employ \emph{homogeneous} MoE-LoRA architectures composed of LoRA experts with either similar or identical structures and capacities. However, these approaches often suffer from representation collapse and expert load imbalance, which negatively impact the potential of LLMs. To address these challenges, we propose a \emph{heterogeneous} \textbf{Mixture-of-Adapters (MoA)} approach. This method dynamically integrates PEFT adapter experts with diverse structures, leveraging their complementary representational capabilities to foster expert specialization, thereby enhancing the effective transfer of pre-trained knowledge to downstream tasks. MoA supports two variants: \textbf{(i)} \textit{Soft MoA} achieves fine-grained integration by performing a weighted fusion of all expert outputs; \textbf{(ii)} \textit{Sparse MoA} activates adapter experts sparsely based on their contribution, achieving this with negligible performance degradation. Experimental results demonstrate that heterogeneous MoA outperforms homogeneous MoE-LoRA methods in both performance and parameter efficiency. Our project is available at https://github.com/DCDmllm/MoA.
Authors: Yixuan Zhu, Haolin Wang, Shilin Ma, Wenliang Zhao, Yansong Tang, Lei Chen, Jie Zhou
Abstract: Recent advancements in diffusion frameworks have significantly enhanced video editing, achieving high fidelity and strong alignment with textual prompts. However, conventional approaches using image diffusion models fall short in handling video dynamics, particularly for challenging temporal edits like motion adjustments. While current video diffusion models produce high-quality results, adapting them for efficient editing remains difficult due to the heavy computational demands that prevent the direct application of previous image editing techniques. To overcome these limitations, we introduce FADE, a training-free yet highly effective video editing approach that fully leverages the inherent priors from pre-trained video diffusion models via frequency-aware factorization. Rather than simply using these models, we first analyze the attention patterns within the video model to reveal how video priors are distributed across different components. Building on these insights, we propose a factorization strategy to optimize each component's specialized role. Furthermore, we devise spectrum-guided modulation to refine the sampling trajectory with frequency domain cues, preventing information leakage and supporting efficient, versatile edits while preserving the basic spatial and temporal structure. Extensive experiments on real-world videos demonstrate that our method consistently delivers high-quality, realistic and temporally coherent editing results both qualitatively and quantitatively. Code is available at https://github.com/EternalEvan/FADE .
Authors: Wei Li, Yanbin Wei, Qiushi Huang, Jiangyue Yan, Yang Chen, James T. Kwok, Yu Zhang
Abstract: Modern large language models (LLMs) often struggle to dynamically adapt their reasoning depth to varying task complexities, leading to suboptimal performance or inefficient resource utilization. To address this, we introduce DynamicMind, a novel tri-mode thinking system. DynamicMind empowers LLMs to autonomously select between Fast, Normal, and Slow thinking modes for zero-shot question answering (ZSQA) tasks through cognitive-inspired prompt engineering. Our framework's core innovations include: (1) expanding the established dual-process framework of fast and slow thinking into a tri-mode thinking system involving a normal thinking mode to preserve the intrinsic capabilities of LLM; (2) proposing the Thinking Density metric, which aligns computational resource allocation with problem complexity; and (3) developing the Thinking Mode Capacity (TMC) dataset and a lightweight Mind Router to predict the optimal thinking mode. Extensive experiments across diverse mathematical, commonsense, and scientific QA benchmarks demonstrate that DynamicMind achieves superior ZSQA capabilities while establishing an effective trade-off between performance and computational efficiency.
Authors: Charmaine Barker, Daniel Bethell, Simos Gerasimou
Abstract: Reliability of deep learning models is critical for deployment in high-stakes applications, where out-of-distribution or adversarial inputs may lead to detrimental outcomes. Evidential Deep Learning, an efficient paradigm for uncertainty quantification, models predictions as Dirichlet distributions of a single forward pass. However, EDL is particularly vulnerable to adversarially perturbed inputs, making overconfident errors. Conflict-aware Evidential Deep Learning (C-EDL) is a lightweight post-hoc uncertainty quantification approach that mitigates these issues, enhancing adversarial and OOD robustness without retraining. C-EDL generates diverse, task-preserving transformations per input and quantifies representational disagreement to calibrate uncertainty estimates when needed. C-EDL's conflict-aware prediction adjustment improves detection of OOD and adversarial inputs, maintaining high in-distribution accuracy and low computational overhead. Our experimental evaluation shows that C-EDL significantly outperforms state-of-the-art EDL variants and competitive baselines, achieving substantial reductions in coverage for OOD data (up to 55%) and adversarial data (up to 90%), across a range of datasets, attack types, and uncertainty metrics.
Authors: Luka Hobor, Mario Brcic, Lidija Polutnik, Ante Kapetanovic
Abstract: Accurate forecasting is key for all business planning. When estimated sales are too high, brick-and-mortar retailers may incur higher costs due to unsold inventories, higher labor and storage space costs, etc. On the other hand, when forecasts underestimate the level of sales, firms experience lost sales, shortages, and impact on the reputation of the retailer in their relevant market. Accurate forecasting presents a competitive advantage for companies. It facilitates the achievement of revenue and profit goals and execution of pricing strategy and tactics. In this study, we provide an exhaustive assessment of the forecasting models applied to a high-resolution brick-and-mortar retail dataset. Our forecasting framework addresses the problems found in retail environments, including intermittent demand, missing values, and frequent product turnover. We compare tree-based ensembles (such as XGBoost and LightGBM) and state-of-the-art neural network architectures (including N-BEATS, NHITS, and the Temporal Fusion Transformer) across various experimental settings. Our results show that localized modeling strategies especially those using tree-based models on individual groups with non-imputed data, consistently deliver superior forecasting accuracy and computational efficiency. In contrast, neural models benefit from advanced imputation methods, yet still fall short in handling the irregularities typical of physical retail data. These results further practical understanding for model selection in retail environment and highlight the significance of data preprocessing to improve forecast performance.
Authors: Xinjie Zhang, Wenxuan Wang, Qin Jin
Abstract: In emotional support conversations, unclear intentions can lead supporters to employ inappropriate strategies, inadvertently imposing their expectations or solutions on the seeker. Clearly defined intentions are essential for guiding both the supporter's motivations and the overall emotional support process. In this paper, we propose the Intention-centered Emotional Support Conversation (IntentionESC) framework, which defines the possible intentions of supporters in emotional support conversations, identifies key emotional state aspects for inferring these intentions, and maps them to appropriate support strategies. While Large Language Models (LLMs) excel in text generating, they fundamentally operate as probabilistic models trained on extensive datasets, lacking a true understanding of human thought processes and intentions. To address this limitation, we introduce the Intention Centric Chain-of-Thought (ICECoT) mechanism. ICECoT enables LLMs to mimic human reasoning by analyzing emotional states, inferring intentions, and selecting suitable support strategies, thereby generating more effective emotional support responses. To train the model with ICECoT and integrate expert knowledge, we design an automated annotation pipeline that produces high-quality training data. Furthermore, we develop a comprehensive evaluation scheme to assess emotional support efficacy and conduct extensive experiments to validate our framework. Our data and code are available at https://github.com/43zxj/IntentionESC_ICECoT.
Authors: Dongjie Fu, Tengjiao Sun, Pengcheng Fang, Xiaohao Cai, Hansung Kim
Abstract: Recent advances in transformer-based text-to-motion generation have led to impressive progress in synthesizing high-quality human motion. Nevertheless, jointly achieving high fidelity, streaming capability, real-time responsiveness, and scalability remains a fundamental challenge. In this paper, we propose MOGO (Motion Generation with One-pass), a novel autoregressive framework tailored for efficient and real-time 3D motion generation. MOGO comprises two key components: (1) MoSA-VQ, a motion scale-adaptive residual vector quantization module that hierarchically discretizes motion sequences with learnable scaling to produce compact yet expressive representations; and (2) RQHC-Transformer, a residual quantized hierarchical causal transformer that generates multi-layer motion tokens in a single forward pass, significantly reducing inference latency. To enhance semantic fidelity, we further introduce a text condition alignment mechanism that improves motion decoding under textual control. Extensive experiments on benchmark datasets including HumanML3D, KIT-ML, and CMP demonstrate that MOGO achieves competitive or superior generation quality compared to state-of-the-art transformer-based methods, while offering substantial improvements in real-time performance, streaming generation, and generalization under zero-shot settings.
Authors: Motoki Omura, Kazuki Ota, Takayuki Osa, Yusuke Mukuta, Tatsuya Harada
Abstract: For continuous action spaces, actor-critic methods are widely used in online reinforcement learning (RL). However, unlike RL algorithms for discrete actions, which generally model the optimal value function using the Bellman optimality operator, RL algorithms for continuous actions typically model Q-values for the current policy using the Bellman operator. These algorithms for continuous actions rely exclusively on policy updates for improvement, which often results in low sample efficiency. This study examines the effectiveness of incorporating the Bellman optimality operator into actor-critic frameworks. Experiments in a simple environment show that modeling optimal values accelerates learning but leads to overestimation bias. To address this, we propose an annealing approach that gradually transitions from the Bellman optimality operator to the Bellman operator, thereby accelerating learning while mitigating bias. Our method, combined with TD3 and SAC, significantly outperforms existing approaches across various locomotion and manipulation tasks, demonstrating improved performance and robustness to hyperparameters related to optimality.
Authors: Kazutoshi Shinoda, Nobukatsu Hojo, Kyosuke Nishida, Yoshihiro Yamazaki, Keita Suzuki, Hiroaki Sugiyama, Kuniko Saito
Abstract: Recent studies have shown that Theory of Mind (ToM) in large language models (LLMs) has not reached human-level performance yet. Since fine-tuning LLMs on ToM datasets often degrades their generalization, several inference-time methods have been proposed to enhance ToM in LLMs. However, existing inference-time methods for ToM are specialized for inferring beliefs from contexts involving changes in the world state. In this study, we present a new inference-time method for ToM, Shoes-of-Others (SoO) prefixing, which makes fewer assumptions about contexts and is applicable to broader scenarios. SoO prefixing simply specifies the beginning of LLM outputs with ``Let's put ourselves in A's shoes.'', where A denotes the target character's name. We evaluate SoO prefixing on two benchmarks that assess ToM in conversational and narrative contexts without changes in the world state and find that it consistently improves ToM across five categories of mental states. Our analysis suggests that SoO prefixing elicits faithful thoughts, thereby improving the ToM performance.
Authors: Jacob Bamberger, Benjamin Gutteridge, Scott le Roux, Michael M. Bronstein, Xiaowen Dong
Abstract: Long-range graph tasks -- those dependent on interactions between distant nodes -- are an open problem in graph neural network research. Real-world benchmark tasks, especially the Long Range Graph Benchmark, have become popular for validating the long-range capability of proposed architectures. However, this is an empirical approach that lacks both robustness and theoretical underpinning; a more principled characterization of the long-range problem is required. To bridge this gap, we formalize long-range interactions in graph tasks, introduce a range measure for operators on graphs, and validate it with synthetic experiments. We then leverage our measure to examine commonly used tasks and architectures, and discuss to what extent they are, in fact, long-range. We believe our work advances efforts to define and address the long-range problem on graphs, and that our range measure will aid evaluation of new datasets and architectures.
Authors: Geonwoo Cho, Jaemoon Lee, Jaegyun Im, Subi Lee, Jihwan Lee, Sundong Kim
Abstract: Skill-based reinforcement learning (SBRL) enables rapid adaptation in environments with sparse rewards by pretraining a skill-conditioned policy. Effective skill learning requires jointly maximizing both exploration and skill diversity. However, existing methods often face challenges in simultaneously optimizing for these two conflicting objectives. In this work, we propose a new method, Adaptive Multi-objective Projection for balancing Exploration and skill Diversification (AMPED), which explicitly addresses both exploration and skill diversification. We begin by conducting extensive ablation studies to identify and define a set of objectives that effectively capture the aspects of exploration and skill diversity, respectively. During the skill pretraining phase, AMPED introduces a gradient surgery technique to balance the objectives of exploration and skill diversity, mitigating conflicts and reducing reliance on heuristic tuning. In the subsequent fine-tuning phase, AMPED incorporates a skill selector module that dynamically selects suitable skills for downstream tasks, based on task-specific performance signals. Our approach achieves performance that surpasses SBRL baselines across various benchmarks. These results highlight the importance of explicitly harmonizing exploration and diversity and demonstrate the effectiveness of AMPED in enabling robust and generalizable skill learning. Project Page: https://geonwoo.me/amped/
Authors: Cheng-Han Chiang, Xiaofei Wang, Chung-Ching Lin, Kevin Lin, Linjie Li, Radu Kopetz, Yao Qian, Zhendong Wang, Zhengyuan Yang, Hung-yi Lee, Lijuan Wang
Abstract: Audio-aware large language models (ALLMs) can understand the textual and non-textual information in the audio input. In this paper, we explore using ALLMs as an automatic judge to assess the speaking styles of speeches. We use ALLM judges to evaluate the speeches generated by SLMs on two tasks: voice style instruction following and role-playing. The speaking style we consider includes emotion, volume, speaking pace, word emphasis, pitch control, and non-verbal elements. We use four spoken language models (SLMs) to complete the two tasks and use humans and ALLMs to judge the SLMs' responses. We compare two ALLM judges, GPT-4o-audio and Gemini-2.5-pro, with human evaluation results and show that the agreement between Gemini and human judges is comparable to the agreement between human evaluators. These promising results show that ALLMs can be used as a judge to evaluate SLMs. Our results also reveal that current SLMs, even GPT-4o-audio, still have room for improvement in controlling the speaking style and generating natural dialogues.
Authors: Stefan Dascalescu, Adrian Marius Dumitran, Mihai Alexandru Vasiluta
Abstract: Competitive programming contests play a crucial role in cultivating computational thinking and algorithmic skills among learners. However, generating comprehensive test cases to effectively assess programming solutions remains resource-intensive and challenging for educators. This paper introduces an innovative NLP-driven method leveraging generative AI (large language models) to automate the creation of high-quality test cases for competitive programming assessments. We extensively evaluated our approach on diverse datasets, including 25 years of Romanian Informatics Olympiad (OJI) data for 5th graders, recent competitions hosted on the Kilonova.ro platform, and the International Informatics Olympiad in Teams (IIOT). Our results demonstrate that AI-generated test cases substantially enhanced assessments, notably identifying previously undetected errors in 67% of the OJI 5th grade programming problems. These improvements underscore the complementary educational value of our technique in formative assessment contexts. By openly sharing our prompts, translated datasets, and methodologies, we offer practical NLP-based tools that educators and contest organizers can readily integrate to enhance assessment quality, reduce workload, and deepen insights into learner performance.
Authors: Yifu Qiu, Yftah Ziser, Anna Korhonen, Shay B. Cohen, Edoardo M. Ponti
Abstract: To what extent do vision-and-language foundation models possess a realistic world model (observation $\times$ action $\rightarrow$ observation) and a dynamics model (observation $\times$ observation $\rightarrow$ action), when actions are expressed through language? While open-source foundation models struggle with both, we find that fine-tuning them to acquire a dynamics model through supervision is significantly easier than acquiring a world model. In turn, dynamics models can be used to bootstrap world models through two main strategies: 1) weakly supervised learning from synthetic data and 2) inference time verification. Firstly, the dynamics model can annotate actions for unlabelled pairs of video frame observations to expand the training data. We further propose a new objective, where image tokens in observation pairs are weighted by their importance, as predicted by a recognition model. Secondly, the dynamics models can assign rewards to multiple samples of the world model to score them, effectively guiding search at inference time. We evaluate the world models resulting from both strategies through the task of action-centric image editing on Aurora-Bench. Our best model achieves a performance competitive with state-of-the-art image editing models, improving on them by a margin of $15\%$ on real-world subsets according to GPT4o-as-judge, and achieving the best average human evaluation across all subsets of Aurora-Bench.
Authors: Alejandro Puente-Castro, Enrique Fernandez-Blanco, Daniel Rivero, Andres Molares-Ulloa
Abstract: Orthopoxvirus infections must be accurately classified from medical pictures for an easy and early diagnosis and epidemic prevention. The necessity for automated and scalable solutions is highlighted by the fact that traditional diagnostic techniques can be time-consuming and require expert interpretation and there are few and biased data sets of the different types of Orthopox. In order to improve classification performance and lower computational costs, a hybrid strategy is put forth in this paper that uses Machine Learning models combined with pretrained Deep Learning models to extract deep feature representations without the need for augmented data. The findings show that this feature extraction method, when paired with other methods in the state-of-the-art, produces excellent classification outcomes while preserving training and inference efficiency. The proposed approach demonstrates strong generalization and robustness across multiple evaluation settings, offering a scalable and interpretable solution for real-world clinical deployment.
Authors: Peijie Liu, Fengli Xu, Yong Li
Abstract: Chain-of-Thought (CoT) technique has proven effective in improving the performance of large language models (LLMs) on complex reasoning tasks. However, the performance gains are inconsistent across different tasks, and the underlying mechanism remains a long-standing research question. In this work, we make a preliminary observation that the monotonicity of token probability distributions may be correlated with the gains achieved through CoT reasoning. Leveraging this insight, we propose two indicators based on the token probability distribution to assess CoT effectiveness across different tasks. By combining instance-level indicators with logistic regression model, we introduce Dynamic CoT, a method that dynamically select between CoT and direct answer. Furthermore, we extend Dynamic CoT to closed-source models by transferring decision strategies learned from open-source models. Our indicators for assessing CoT effectiveness achieve an accuracy of 89.2\%, and Dynamic CoT reduces token consumption by more than 35\% while maintaining high accuracy. Overall, our work offers a novel perspective on the underlying mechanisms of CoT reasoning and provides a framework for its more efficient deployment.
Authors: Haoke Zhang, Xiaobo Liang, Cunxiang Wang, Juntao Li, Min Zhang
Abstract: The OpenAI o1-series models have demonstrated that leveraging long-form Chain of Thought (CoT) can substantially enhance performance. However, the recursive thinking capabilities of Large Language Models (LLMs) remain limited, particularly in the absence of expert-curated data for distillation. In this paper, we propose \textbf{AvR}: \textbf{Alignment via Refinement}, a novel method aimed at unlocking the potential of LLMs for recursive reasoning through long-form CoT. AvR introduces a refinement process that integrates criticism and improvement actions, guided by differentiable learning techniques to optimize \textbf{refinement-aware rewards}. As a result, the synthesized multi-round data can be organized as a long refinement thought, further enabling test-time scaling. Experimental results show that AvR significantly outperforms conventional preference optimization methods. Notably, with only 3k synthetic samples, our method boosts the performance of the LLaMA-3-8B-Instruct model by over 20\% in win rate on AlpacaEval 2.0. Our code is available at Github (https://github.com/Banner-Z/AvR.git).
Authors: Chaoyi Zhu, Zaitang Li, Renyi Yang, Robert Birke, Pin-Yu Chen, Tsung-Yi Ho, Lydia Y. Chen
Abstract: Watermarking becomes one of the pivotal solutions to trace and verify the origin of synthetic images generated by artificial intelligence models, but it is not free of risks. Recent studies demonstrate the capability to forge watermarks from a target image onto cover images via adversarial optimization without knowledge of the target generative model and watermark schemes. In this paper, we uncover a greater risk of an optimization-free and universal watermark forgery that harnesses existing regenerative diffusion models. Our proposed forgery attack, PnP (Plug-and-Plant), seamlessly extracts and integrates the target watermark via regenerating the image, without needing any additional optimization routine. It allows for universal watermark forgery that works independently of the target image's origin or the watermarking model used. We explore the watermarked latent extracted from the target image and visual-textual context of cover images as priors to guide sampling of the regenerative process. Extensive evaluation on 24 scenarios of model-data-watermark combinations demonstrates that PnP can successfully forge the watermark (up to 100% detectability and user attribution), and maintain the best visual perception. By bypassing model retraining and enabling adaptability to any image, our approach significantly broadens the scope of forgery attacks, presenting a greater challenge to the security of current watermarking techniques for diffusion models and the authority of watermarking schemes in synthetic data generation and governance.
Authors: Zeqi Zhou, Fang Wu, Shayan Talaei, Haokai Zhao, Cheng Meixin, Tinson Xu, Amin Saberi, Yejin Choi
Abstract: Large language models frequently encounter conflicts between their parametric knowledge and contextual input, often resulting in factual inconsistencies or hallucinations. We propose Self-Reflective Debate for Contextual Reliability (SR-DCR), a lightweight framework that integrates token-level self-confidence with an asymmetric multi-agent debate to adjudicate such conflicts. A critic, deprived of context, challenges a defender who argues from the given passage; a judge model evaluates the debate and determines the context's reliability. The final answer is selected by combining the verdict with model confidence. Experiments on the ClashEval benchmark demonstrate that SR-DCR consistently enhances robustness to misleading context while maintaining accuracy on trustworthy inputs, outperforming both classical debate and confidence-only baselines with minimal computational overhead. The code is available at https://github.com/smiles724/Self-Reflective-Debates.
Authors: Nikunj Shah, Utsav Dey, Kenji Nishimiya
Abstract: Efficient Coverage Path Planning (CPP) is necessary for autonomous robotic lawnmowers to effectively navigate and maintain lawns with diverse and irregular shapes. This paper introduces a comprehensive end-to-end pipeline for CPP, designed to convert user-defined boundaries on an aerial map into optimized coverage paths seamlessly. The pipeline includes user input extraction, coordinate transformation, area decomposition and path generation using our novel AdaptiveDecompositionCPP algorithm, preview and customization through an interactive coverage path visualizer, and conversion to actionable GPS waypoints. The AdaptiveDecompositionCPP algorithm combines cellular decomposition with an adaptive merging strategy to reduce non-mowing travel thereby enhancing operational efficiency. Experimental evaluations, encompassing both simulations and real-world lawnmower tests, demonstrate the effectiveness of the framework in coverage completeness and mowing efficiency.
Authors: Shiyi Zhang, Dong Liang, Hairong Zheng, Yihang Zhou
Abstract: Reconstructing visual information from brain activity bridges the gap between neuroscience and computer vision. Even though progress has been made in decoding images from fMRI using generative models, a challenge remains in accurately recovering highly complex visual stimuli. This difficulty stems from their elemental density and diversity, sophisticated spatial structures, and multifaceted semantic information. To address these challenges, we propose HAVIR that contains two adapters: (1) The AutoKL Adapter transforms fMRI voxels into a latent diffusion prior, capturing topological structures; (2) The CLIP Adapter converts the voxels to CLIP text and image embeddings, containing semantic information. These complementary representations are fused by Versatile Diffusion to generate the final reconstructed image. To extract the most essential semantic information from complex scenarios, the CLIP Adapter is trained with text captions describing the visual stimuli and their corresponding semantic images synthesized from these captions. The experimental results demonstrate that HAVIR effectively reconstructs both structural features and semantic information of visual stimuli even in complex scenarios, outperforming existing models.
Authors: Haripriya Harikumar, Santu Rana
Abstract: Standard uncertainty estimation techniques, such as dropout, often struggle to clearly distinguish reliable predictions from unreliable ones. We attribute this limitation to noisy classifier weights, which, while not impairing overall class-level predictions, render finer-level statistics less informative. To address this, we propose a novel test-time optimization method that accounts for the impact of such noise to produce more reliable confidence estimates. This score defines a monotonic subset-selection function, where population accuracy consistently increases as samples with lower scores are removed, and it demonstrates superior performance in standard risk-based metrics such as AUSE and AURC. Additionally, our method effectively identifies discrepancies between training and test distributions, reliably differentiates in-distribution from out-of-distribution samples, and elucidates key differences between CNN and ViT classifiers across various vision datasets.
Authors: Minglang Chen, Jie He, Caixu Xu, Bocheng Liang, Shengli Li, Guannan He, Xiongjie Tao
Abstract: ResNet has been widely used in image classification tasks due to its ability to model the residual dependence of constant mappings for linear computation. However, the ResNet method adopts a unidirectional transfer of features and lacks an effective method to correlate contextual information, which is not effective in classifying fetal ultrasound images in the classification task, and fetal ultrasound images have problems such as low contrast, high similarity, and high noise. Therefore, we propose a bilateral multi-scale information fusion network-based FPDANet to address the above challenges. Specifically, we design the positional attention mechanism (DAN) module, which utilizes the similarity of features to establish the dependency of different spatial positional features and enhance the feature representation. In addition, we design a bilateral multi-scale (FPAN) information fusion module to capture contextual and global feature dependencies at different feature scales, thereby further improving the model representation. FPDANet classification results obtained 91.05\% and 100\% in Top-1 and Top-5 metrics, respectively, and the experimental results proved the effectiveness and robustness of FPDANet.
Authors: Chen Xiong, Zihao Wang, Rui Zhu, Tsung-Yi Ho, Pin-Yu Chen, Jingwei Xiong, Haixu Tang, Lucila Ohno-Machado
Abstract: Large Language Models (LLMs) have revolutionized Natural Language Processing by excelling at interpreting, reasoning about, and generating human language. However, their reliance on large-scale, often proprietary datasets poses a critical challenge: unauthorized usage of such data can lead to copyright infringement and significant financial harm. Existing dataset-inference methods typically depend on log probabilities to detect suspicious training material, yet many leading LLMs have begun withholding or obfuscating these signals. This reality underscores the pressing need for label-only approaches capable of identifying dataset membership without relying on internal model logits. We address this gap by introducing CatShift, a label-only dataset-inference framework that capitalizes on catastrophic forgetting: the tendency of an LLM to overwrite previously learned knowledge when exposed to new data. If a suspicious dataset was previously seen by the model, fine-tuning on a portion of it triggers a pronounced post-tuning shift in the model's outputs; conversely, truly novel data elicits more modest changes. By comparing the model's output shifts for a suspicious dataset against those for a known non-member validation set, we statistically determine whether the suspicious set is likely to have been part of the model's original training corpus. Extensive experiments on both open-source and API-based LLMs validate CatShift's effectiveness in logit-inaccessible settings, offering a robust and practical solution for safeguarding proprietary data.
Authors: Viorica Chifu, Cristina Bianca Pop, Tudor Cioara, Ionut Anghel
Abstract: With the integration of renewable sources in electricity distribution networks, the need to develop intelligent mechanisms for balancing the energy market has arisen. In the absence of such mechanisms, the energy market may face imbalances that can lead to power outages, financial losses or instability at the grid level. In this context, the grouping of microgrids into optimal coalitions that can absorb energy from the market during periods of surplus or supply energy to the market during periods of is a key aspect in the efficient management of distribution networks. In this article, we propose a method that identify an optimal microgrids coalition capable of addressing the dynamics of the energy market. The proposed method models the problem of identifying the optimal coalition as an optimization problem that it solves by combining a strategy inspired by cooperative game theory with a memetic algorithm. An individual is represented as a coalition of microgrids and the evolution of population of individuals over generations is assured by recombination and mutation. The fitness function is defined as the difference between the total value generated by the coalition and a penalty applied to the coalition when the energy traded by coalition exceeds the energy available/demanded on/by the energy market. The value generated by the coalition is calculated based on the profit obtained by the collation if it sells energy on the market during periods of deficit or the savings obtained by the coalition if it buys energy on the market during periods of surplus and the costs associated with the trading process. This value is divided equitably among the coalition members, according to the Shapley value, which considers the contribution of each one to the formation of collective value.
Authors: Yingqi Hu, Zhuo Zhang, Jingyuan Zhang, Lizhen Qu, Zenglin Xu
Abstract: Federated fine-tuning of large language models (FedLLMs) presents a promising approach for achieving strong model performance while preserving data privacy in sensitive domains. However, the inherent memorization ability of LLMs makes them vulnerable to training data extraction attacks. To investigate this risk, we introduce simple yet effective extraction attack algorithms specifically designed for FedLLMs. In contrast to prior "verbatim" extraction attacks, which assume access to fragments from all training data, our approach operates under a more realistic threat model, where the attacker only has access to a single client's data and aims to extract previously unseen personally identifiable information (PII) from other clients. This requires leveraging contextual prefixes held by the attacker to generalize across clients. To evaluate the effectiveness of our approaches, we propose two rigorous metrics-coverage rate and efficiency-and extend a real-world legal dataset with PII annotations aligned with CPIS, GDPR, and CCPA standards, achieving 89.9% human-verified precision. Experimental results show that our method can extract up to 56.57% of victim-exclusive PII, with "Address," "Birthday," and "Name" being the most vulnerable categories. Our findings underscore the pressing need for robust defense strategies and contribute a new benchmark and evaluation framework for future research in privacy-preserving federated learning.
Authors: Rujikorn Charakorn, Edoardo Cetin, Yujin Tang, Robert Tjarko Lange
Abstract: While Foundation Models provide a general tool for rapid content creation, they regularly require task-specific adaptation. Traditionally, this exercise involves careful curation of datasets and repeated fine-tuning of the underlying model. Fine-tuning techniques enable practitioners to adapt foundation models for many new applications but require expensive and lengthy training while being notably sensitive to hyper-parameter choices. To overcome these limitations, we introduce Text-to-LoRA (T2L), a model capable of adapting Large Language Models on the fly solely based on a natural language description of the target task. T2L is a hypernetwork trained to construct LoRAs in a single inexpensive forward pass. After training T2L on a suite of 9 pre-trained LoRA adapters (GSM8K, Arc, etc.), we show that the ad-hoc reconstructed LoRA instances match the performance of task-specific adapters across the corresponding test sets. Furthermore, T2L can compress hundreds of LoRA instances and zero-shot generalize to entirely unseen tasks. This approach provides a significant step towards democratizing the specialization of foundation models and enables language-based adaptation with minimal compute requirements. Our code is available at https://github.com/SakanaAI/text-to-lora
Authors: Cheng-Long Wang, Qi Li, Zihang Xiang, Yinzhi Cao, Di Wang
Abstract: Growing concerns over data privacy and security highlight the importance of machine unlearning--removing specific data influences from trained models without full retraining. Techniques like Membership Inference Attacks (MIAs) are widely used to externally assess successful unlearning. However, existing methods face two key limitations: (1) maximizing MIA effectiveness (e.g., via online attacks) requires prohibitive computational resources, often exceeding retraining costs; (2) MIAs, designed for binary inclusion tests, struggle to capture granular changes in approximate unlearning. To address these challenges, we propose the Interpolated Approximate Measurement (IAM), a framework natively designed for unlearning inference. IAM quantifies sample-level unlearning completeness by interpolating the model's generalization-fitting behavior gap on queried samples. IAM achieves strong performance in binary inclusion tests for exact unlearning and high correlation for approximate unlearning--scalable to LLMs using just one pre-trained shadow model. We theoretically analyze how IAM's scoring mechanism maintains performance efficiently. We then apply IAM to recent approximate unlearning algorithms, revealing general risks of both over-unlearning and under-unlearning, underscoring the need for stronger safeguards in approximate unlearning systems. The code is available at https://github.com/Happy2Git/Unlearning_Inference_IAM.
URLs: https://github.com/Happy2Git/Unlearning_Inference_IAM.
Authors: Christophe Van Gysel, Maggie Wu, Lyan Verwimp, Caglar Tirkaz, Marco Bertola, Zhihong Lei, Youssef Oualil
Abstract: End-to-end (E2E) Automatic Speech Recognition (ASR) models are trained using paired audio-text samples that are expensive to obtain, since high-quality ground-truth data requires human annotators. Voice search applications, such as digital media players, leverage ASR to allow users to search by voice as opposed to an on-screen keyboard. However, recent or infrequent movie titles may not be sufficiently represented in the E2E ASR system's training data, and hence, may suffer poor recognition. In this paper, we propose a phonetic correction system that consists of (a) a phonetic search based on the ASR model's output that generates phonetic alternatives that may not be considered by the E2E system, and (b) a rescorer component that combines the ASR model recognition and the phonetic alternatives, and select a final system output. We find that our approach improves word error rate between 4.4 and 7.6% relative on benchmarks of popular movie titles over a series of competitive baselines.
Authors: Haowei Wang, Rupeng Zhang, Junjie Wang, Mingyang Li, Yuekai Huang, Dandan Wang, Qing Wang
Abstract: Retrieval-Augmented Generation (RAG) systems enhance Large Language Models (LLMs) by retrieving relevant documents from external corpora before generating responses. This approach significantly expands LLM capabilities by leveraging vast, up-to-date external knowledge. However, this reliance on external knowledge makes RAG systems vulnerable to corpus poisoning attacks that manipulate generated outputs via poisoned document injection. Existing poisoning attack strategies typically treat the retrieval and generation stages as disjointed, limiting their effectiveness. We propose Joint-GCG, the first framework to unify gradient-based attacks across both retriever and generator models through three innovations: (1) Cross-Vocabulary Projection for aligning embedding spaces, (2) Gradient Tokenization Alignment for synchronizing token-level gradient signals, and (3) Adaptive Weighted Fusion for dynamically balancing attacking objectives. Evaluations demonstrate that Joint-GCG achieves at most 25% and an average of 5% higher attack success rate than previous methods across multiple retrievers and generators. While optimized under a white-box assumption, the generated poisons show unprecedented transferability to unseen models. Joint-GCG's innovative unification of gradient-based attacks across retrieval and generation stages fundamentally reshapes our understanding of vulnerabilities within RAG systems. Our code is available at https://github.com/NicerWang/Joint-GCG.
Authors: Zackary Okun Dunivin, Paul E. Smaldino
Abstract: Scientific recommender systems, such as Google Scholar and Web of Science, are essential tools for discovery. Search algorithms that power work through stigmergy, a collective intelligence mechanism that surfaces useful paths through repeated engagement. While generally effective, this ``rich-get-richer'' dynamic results in a small number of high-profile papers that dominate visibility. This essay argues argue that these algorithm over-reliance on popularity fosters intellectual homogeneity and exacerbates structural inequities, stifling innovative and diverse perspectives critical for scientific progress. We propose an overhaul of search platforms to incorporate user-specific calibration, allowing researchers to manually adjust the weights of factors like popularity, recency, and relevance. We also advise platform developers on how word embeddings and LLMs could be implemented in ways that increase user autonomy. While our suggestions are particularly pertinent to aligning recommender systems with scientific values, these ideas are broadly applicable to information access systems in general. Designing platforms that increase user autonomy is an important step toward more robust and dynamic information
Authors: Eunhye Grace Ko, Soo Hyoung Joo
Abstract: While the complexity of 21st-century demands has promoted pedagogical approaches to foster complex competencies, a persistent gap remains between in-class learning activities and individualized learning or assessment practices. To address this, studies have explored the use of AI-generated characters in learning and assessment. One attempt is scenario-based assessment (SBA), a technique that not only measures but also fosters the development of competencies throughout the assessment process. SBA introduces simulated agents to provide an authentic social-interactional context, allowing for the assessment of competency-based constructs while mitigating the unpredictability of real-life interactions. Recent advancements in multimodal AI, such as text-to-video technology, allow these agents to be enhanced into AI-generated characters. This mixed-method study investigates how learners perceive AI characters taking the role of mentor and teammates in an SBA mirroring the context of a collaborative science investigation. Specifically, we examined the Likert scale responses of 56 high schoolers regarding trust, social presence, and effectiveness. We analyzed the relationships between these factors and their impact on the intention to adopt AI characters through PLS-SEM. Our findings indicated that learners' trust shaped their sense of social presence with the AI characters, enhancing perceived effectiveness. Qualitative analysis further highlighted factors that foster trust, such as material credibility and alignment with learning goals, as well as the pivotal role of social presence in creating a collaborative context. This paper was accepted as an full paper for AIED 2025.
Authors: Tianyi Alex Qiu, Zhonghao He, Tejasveer Chugh, Max Kleiman-Weiner
Abstract: The training and deployment of large language models (LLMs) create a feedback loop with human users: models learn human beliefs from data, reinforce these beliefs with generated content, reabsorb the reinforced beliefs, and feed them back to users again and again. This dynamic resembles an echo chamber. We hypothesize that this feedback loop entrenches the existing values and beliefs of users, leading to a loss of diversity and potentially the lock-in of false beliefs. We formalize this hypothesis and test it empirically with agent-based LLM simulations and real-world GPT usage data. Analysis reveals sudden but sustained drops in diversity after the release of new GPT iterations, consistent with the hypothesized human-AI feedback loop. Code and data available at https://thelockinhypothesis.com
Authors: Jwalanthi Ranganathan, Rohan Jha, Kanishka Misra, Kyle Mahowald
Abstract: We introduce semantic-features, an extensible, easy-to-use library based on Chronis et al. (2023) for studying contextualized word embeddings of LMs by projecting them into interpretable spaces. We apply this tool in an experiment where we measure the contextual effect of the choice of dative construction (prepositional or double object) on the semantic interpretation of utterances (Bresnan, 2007). Specifically, we test whether "London" in "I sent London the letter." is more likely to be interpreted as an animate referent (e.g., as the name of a person) than in "I sent the letter to London." To this end, we devise a dataset of 450 sentence pairs, one in each dative construction, with recipients being ambiguous with respect to person-hood vs. place-hood. By applying semantic-features, we show that the contextualized word embeddings of three masked language models show the expected sensitivities. This leaves us optimistic about the usefulness of our tool.
Authors: Renato Cordeiro Ferreira (Jheronimus Academy of Data Science, Technical University of Eindhoven, Tilburg University), Rowanne Trapmann (Jheronimus Academy of Data Science, Technical University of Eindhoven, Tilburg University), Willem-Jan van den Heuvel (Jheronimus Academy of Data Science, Technical University of Eindhoven, Tilburg University)
Abstract: This case study describes challenges and lessons learned on building Ocean Guard: a Machine Learning-Enabled System (MLES) for anomaly detection in the maritime domain. First, the paper presents the system's specification, and architecture. Ocean Guard was designed with a microservices' architecture to enable multiple teams to work on the project in parallel. Then, the paper discusses how the developers adapted contract-based design to MLOps for achieving that goal. As a MLES, Ocean Guard employs code, model, and data contracts to establish guidelines between its services. This case study hopes to inspire software engineers, machine learning engineers, and data scientists to leverage similar approaches for their systems.
Authors: Sheng Chen, Peiyu He, Jiaxin Hu, Ziyang Liu, Yansheng Wang, Tao Xu, Chi Zhang, Chongchong Zhang, Chao An, Shiyu Cai, Duo Cao, Kangping Chen, Shuai Chu, Tianwei Chu, Mingdi Dan, Min Du, Weiwei Fang, Pengyou Fu, Junkai Hu, Xiaowei Jiang, Zhaodi Jiang, Fuxuan Li, Jun Li, Minghui Li, Mingyao Li, Yanchang Li, Zhibin Li, Guangming Liu, Kairui Liu, Lihao Liu, Weizhi Liu, Xiaoshun Liu, Yufei Liu, Yunfei Liu, Qiang Lu, Yuanfei Luo, Xiang Lv, Hongying Ma, Sai Ma, Lingxian Mi, Sha Sa, Hongxiang Shu, Lei Tian, Chengzhi Wang, Jiayu Wang, Kaijie Wang, Qingyi Wang, Renwen Wang, Tao Wang, Wei Wang, Xirui Wang, Chao Wei, Xuguang Wei, Zijun Xia, Zhaohao Xiao, Tingshuai Yan, Liyan Yang, Yifan Yang, Zhikai Yang, Zhong Yin, Li Yuan, Liuchun Yuan, Chi Zhang, Jinyang Zhang, Junhui Zhang, Linge Zhang, Zhenyi Zhang, Zheyu Zhang, Dongjie Zhu, Hang Li, Yangang Zhang
Abstract: Modern robot navigation systems encounter difficulties in diverse and complex indoor environments. Traditional approaches rely on multiple modules with small models or rule-based systems and thus lack adaptability to new environments. To address this, we developed Astra, a comprehensive dual-model architecture, Astra-Global and Astra-Local, for mobile robot navigation. Astra-Global, a multimodal LLM, processes vision and language inputs to perform self and goal localization using a hybrid topological-semantic graph as the global map, and outperforms traditional visual place recognition methods. Astra-Local, a multitask network, handles local path planning and odometry estimation. Its 4D spatial-temporal encoder, trained through self-supervised learning, generates robust 4D features for downstream tasks. The planning head utilizes flow matching and a novel masked ESDF loss to minimize collision risks for generating local trajectories, and the odometry head integrates multi-sensor inputs via a transformer encoder to predict the relative pose of the robot. Deployed on real in-house mobile robots, Astra achieves high end-to-end mission success rate across diverse indoor environments.
Authors: Henry Watkins
Abstract: This report documents the development and evaluation of domain-specific language models for neurology. Initially focused on building a bespoke model, the project adapted to rapid advances in open-source and commercial medical LLMs, shifting toward leveraging retrieval-augmented generation (RAG) and representational models for secure, local deployment. Key contributions include the creation of neurology-specific datasets (case reports, QA sets, textbook-derived data), tools for multi-word expression extraction, and graph-based analyses of medical terminology. The project also produced scripts and Docker containers for local hosting. Performance metrics and graph community results are reported, with future possible work open for multimodal models using open-source architectures like phi-4.
Authors: Hengzhi Li, Brendon Jiang, Alexander Naehu, Regan Song, Justin Zhang, Megan Tjandrasuwita, Chanakya Ekbote, Steven-Shine Chen, Adithya Balachandran, Wei Dai, Rebecca Chang, Paul Pu Liang
Abstract: Puzzlehunts are a genre of complex, multi-step puzzles lacking well-defined problem definitions. In contrast to conventional reasoning benchmarks consisting of tasks with clear instructions, puzzlehunts require models to discover the underlying problem structure from multimodal evidence and iterative reasoning, mirroring real-world domains such as scientific discovery, exploratory data analysis, or investigative problem-solving. Despite recent progress in foundation models, their performance on such open-ended settings remains largely untested. In this paper, we introduce PuzzleWorld, a large-scale benchmark of 667 puzzlehunt-style problems designed to assess step-by-step, open-ended, and creative multimodal reasoning. Each puzzle is annotated with the final solution, detailed reasoning traces, and cognitive skill labels, enabling holistic benchmarking and fine-grained diagnostic analysis. Most state-of-the-art models achieve only 1-2% final answer accuracy, with the best model solving only 14% of puzzles and reaching 40% stepwise accuracy. To demonstrate the value of our reasoning annotations, we show that fine-tuning a small model on reasoning traces improves stepwise reasoning from 4% to 11%, while training on final answers alone degrades performance to near zero. Our error analysis reveals that current models exhibit myopic reasoning, are bottlenecked by the limitations of language-based inference, and lack sketching capabilities crucial for visual and spatial reasoning. We release PuzzleWorld at https://github.com/MIT-MI/PuzzleWorld to support future work on building more general, open-ended, and creative reasoning systems.
Authors: Sirui Lu, Zhijing Jin, Terry Jingchen Zhang, Pavel Kos, J. Ignacio Cirac, Bernhard Sch\"olkopf
Abstract: Large Language Models (LLMs) are rapidly advancing across diverse domains, yet their application in theoretical physics research is not yet mature. This position paper argues that LLM agents can potentially help accelerate theoretical, computational, and applied physics when properly integrated with domain knowledge and toolbox. We analyze current LLM capabilities for physics -- from mathematical reasoning to code generation -- identifying critical gaps in physical intuition, constraint satisfaction, and reliable reasoning. We envision future physics-specialized LLMs that could handle multimodal data, propose testable hypotheses, and design experiments. Realizing this vision requires addressing fundamental challenges: ensuring physical consistency, and developing robust verification methods. We call for collaborative efforts between physics and AI communities to help advance scientific discovery in physics.
Authors: Diji Yang, Minghao Liu, Chung-Hsiang Lo, Yi Zhang, James Davis
Abstract: Vision-language models (VLMs) have shown strong performance on text-to-image retrieval benchmarks. However, bridging this success to real-world applications remains a challenge. In practice, human search behavior is rarely a one-shot action. Instead, it is often a multi-round process guided by clues in mind, that is, a mental image ranging from vague recollections to vivid mental representations of the target image. Motivated by this gap, we study the task of Mental Image Retrieval (MIR), which targets the realistic yet underexplored setting where users refine their search for a mentally envisioned image through multi-round interactions with an image search engine. Central to successful interactive retrieval is the capability of machines to provide users with clear, actionable feedback; however, existing methods rely on indirect or abstract verbal feedback, which can be ambiguous, misleading, or ineffective for users to refine the query. To overcome this, we propose GenIR, a generative multi-round retrieval paradigm leveraging diffusion-based image generation to explicitly reify the AI system's understanding at each round. These synthetic visual representations provide clear, interpretable feedback, enabling users to refine their queries intuitively and effectively. We further introduce a fully automated pipeline to generate a high-quality multi-round MIR dataset. Experimental results demonstrate that GenIR significantly outperforms existing interactive methods in the MIR scenario. This work establishes a new task with a dataset and an effective generative retrieval method, providing a foundation for future research in this direction.
Authors: Eunhye Grace Ko, Shaini Nanayakkara, Earl W. Huff Jr
Abstract: As generative AI (GenAI) becomes increasingly widespread, it is crucial to equip users, particularly vulnerable populations such as older adults (65 and older), with the knowledge to understand its benefits and potential risks. Older adults often exhibit greater reservations about adopting emerging technologies and require tailored literacy support. Using a mixed methods approach, this study examines strategies for delivering GenAI literacy to older adults through a chatbot named Litti, evaluating its impact on their AI literacy (knowledge, safety, and ethical use). The quantitative data indicated a trend toward improved AI literacy, though the results were not statistically significant. However, qualitative interviews revealed diverse levels of familiarity with generative AI and a strong desire to learn more. Findings also show that while Litti provided a positive learning experience, it did not significantly enhance participants' trust or sense of safety regarding GenAI. This exploratory case study highlights the challenges and opportunities in designing AI literacy education for the rapidly growing older adult population.
Authors: Mohammad Jalali, Bahar Dibaei Nia, Farzan Farnia
Abstract: While several feature embedding models have been developed in the literature, comparisons of these embeddings have largely focused on their numerical performance in classification-related downstream applications. However, an interpretable comparison of different embeddings requires identifying and analyzing mismatches between sample groups clustered within the embedding spaces. In this work, we propose the \emph{Spectral Pairwise Embedding Comparison (SPEC)} framework to compare embeddings and identify their differences in clustering a reference dataset. Our approach examines the kernel matrices derived from two embeddings and leverages the eigendecomposition of the difference kernel matrix to detect sample clusters that are captured differently by the two embeddings. We present a scalable implementation of this kernel-based approach, with computational complexity that grows linearly with the sample size. Furthermore, we introduce an optimization problem using this framework to align two embeddings, ensuring that clusters identified in one embedding are also captured in the other model. We provide numerical results demonstrating the SPEC's application to compare and align embeddings on large-scale datasets such as ImageNet and MS-COCO. The code is available at [https://github.com/mjalali/embedding-comparison](github.com/mjalali/embedding-comparison).
URLs: https://github.com/mjalali/embedding-comparison](github.com/mjalali/embedding-comparison).
Authors: Zahra Babaiee, Peyman M. Kiasari, Daniela Rus, Radu Grosu
Abstract: Recent advancements in multimodal large language models have driven breakthroughs in visual question answering. Yet, a critical gap persists, `conceptualization'-the ability to recognize and reason about the same concept despite variations in visual form, a basic ability of human reasoning. To address this challenge, we introduce the Visual Graph Arena (VGA), a dataset featuring six graph-based tasks designed to evaluate and improve AI systems' capacity for visual abstraction. VGA uses diverse graph layouts (e.g., Kamada-Kawai vs. planar) to test reasoning independent of visual form. Experiments with state-of-the-art vision models and multimodal LLMs reveal a striking divide: humans achieved near-perfect accuracy across tasks, while models totally failed on isomorphism detection and showed limited success in path/cycle tasks. We further identify behavioral anomalies suggesting pseudo-intelligent pattern matching rather than genuine understanding. These findings underscore fundamental limitations in current AI models for visual understanding. By isolating the challenge of representation-invariant reasoning, the VGA provides a framework to drive progress toward human-like conceptualization in AI visual models. The Visual Graph Arena is available at: \href{https://vga.csail.mit.edu/}{vga.csail.mit.edu}
Authors: Jingyu Xiao, Ming Wang, Man Ho Lam, Yuxuan Wan, Junliang Liu, Yintong Huo, Michael R. Lyu
Abstract: Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in automated front-end engineering, e.g., generating UI code from visual designs. However, existing front-end UI code generation benchmarks have the following limitations: (1) While framework-based development becomes predominant in modern front-end programming, current benchmarks fail to incorporate mainstream development frameworks. (2) Existing evaluations focus solely on the UI code generation task, whereas practical UI development involves several iterations, including refining editing, and repairing issues. (3) Current benchmarks employ unidimensional evaluation, lacking investigation into influencing factors like task difficulty, input context variations, and in-depth code-level analysis. To bridge these gaps, we introduce DesignBench, a multi-framework, multi-task evaluation benchmark for assessing MLLMs' capabilities in automated front-end engineering. DesignBench encompasses three widely-used UI frameworks (React, Vue, and Angular) alongside vanilla HTML/CSS, and evaluates on three essential front-end tasks (generation, edit, and repair) in real-world development workflows. DesignBench contains 900 webpage samples spanning over 11 topics, 9 edit types, and 6 issue categories, enabling detailed analysis of MLLM performance across multiple dimensions. Our systematic evaluation reveals critical insights into MLLMs' framework-specific limitations, task-related bottlenecks, and performance variations under different conditions, providing guidance for future research in automated front-end development. Our code and data are available at https://github.com/WebPAI/DesignBench.
Authors: Sabri Eyuboglu, Ryan Ehrlich, Simran Arora, Neel Guha, Dylan Zinsley, Emily Liu, Will Tennien, Atri Rudra, James Zou, Azalia Mirhoseini, Christopher Re
Abstract: Large language models are often used to answer queries grounded in large text corpora (e.g. codebases, legal documents, or chat histories) by placing the entire corpus in the context window and leveraging in-context learning (ICL). Although current models support contexts of 100K-1M tokens, this setup is costly to serve because the memory consumption of the KV cache scales with input length. We explore an alternative: training a smaller KV cache offline on each corpus. At inference time, we load this trained KV cache, which we call a Cartridge, and decode a response. Critically, the cost of training a Cartridge can be amortized across all the queries referencing the same corpus. However, we find that the naive approach of training the Cartridge with next-token prediction on the corpus is not competitive with ICL. Instead, we propose self-study, a training recipe in which we generate synthetic conversations about the corpus and train the Cartridge with a context-distillation objective. We find that Cartridges trained with self-study replicate the functionality of ICL, while being significantly cheaper to serve. On challenging long-context benchmarks, Cartridges trained with self-study match ICL performance while using 38.6x less memory and enabling 26.4x higher throughput. Self-study also extends the model's effective context length (e.g. from 128k to 484k tokens on MTOB) and surprisingly, leads to Cartridges that can be composed at inference time without retraining.
Authors: Bruce W. Lee, Addie Foote, Alex Infanger, Leni Shor, Harish Kamath, Jacob Goldman-Wetzler, Bryce Woodworth, Alex Cloud, Alexander Matt Turner
Abstract: Current LLM unlearning methods are not robust: they can be reverted easily with a few steps of finetuning. This is true even for the idealized unlearning method of training to imitate an oracle model that was never exposed to unwanted information, suggesting that output-based finetuning is insufficient to achieve robust unlearning. In a similar vein, we find that training a randomly initialized student to imitate an unlearned model transfers desired behaviors while leaving undesired capabilities behind. In other words, distillation robustifies unlearning. Building on this insight, we propose Unlearn-Noise-Distill-on-Outputs (UNDO), a scalable method that distills an unlearned model into a partially noised copy of itself. UNDO introduces a tunable tradeoff between compute cost and robustness, establishing a new Pareto frontier on synthetic language and arithmetic tasks. At its strongest setting, UNDO matches the robustness of a model retrained from scratch with perfect data filtering while using only 60-80% of the compute and requiring only 0.01% of the pretraining data to be labeled. We also show that UNDO robustifies unlearning on the more realistic Weapons of Mass Destruction Proxy (WMDP) benchmark. Since distillation is widely used in practice, incorporating an unlearning step beforehand offers a convenient path to robust capability removal.
Authors: Yuanzhe Hu, Kinshuk Goel, Vlad Killiakov, Yaoqing Yang
Abstract: Diagnosing deep neural networks (DNNs) through the eigenspectrum of weight matrices has been an active area of research in recent years. At a high level, eigenspectrum analysis of DNNs involves measuring the heavytailness of the empirical spectral densities (ESD) of weight matrices. It provides insight into how well a model is trained and can guide decisions on assigning better layer-wise training hyperparameters. In this paper, we address a challenge associated with such eigenspectrum methods: the impact of the aspect ratio of weight matrices on estimated heavytailness metrics. We demonstrate that matrices of varying sizes (and aspect ratios) introduce a non-negligible bias in estimating heavytailness metrics, leading to inaccurate model diagnosis and layer-wise hyperparameter assignment. To overcome this challenge, we propose FARMS (Fixed-Aspect-Ratio Matrix Subsampling), a method that normalizes the weight matrices by subsampling submatrices with a fixed aspect ratio. Instead of measuring the heavytailness of the original ESD, we measure the average ESD of these subsampled submatrices. We show that measuring the heavytailness of these submatrices with the fixed aspect ratio can effectively mitigate the aspect ratio bias. We validate our approach across various optimization techniques and application domains that involve eigenspectrum analysis of weights, including image classification in computer vision (CV) models, scientific machine learning (SciML) model training, and large language model (LLM) pruning. Our results show that despite its simplicity, FARMS uniformly improves the accuracy of eigenspectrum analysis while enabling more effective layer-wise hyperparameter assignment in these application domains. In one of the LLM pruning experiments, FARMS reduces the perplexity of the LLaMA-7B model by 17.3% when compared with the state-of-the-art method.
Authors: Matthew L. Daggitt, Wen Kokke, Robert Atkey, Ekaterina Komendantskaya, Natalia Slusarz, Luca Arnaboldi
Abstract: Neuro-symbolic programs, i.e. programs containing both machine learning components and traditional symbolic code, are becoming increasingly widespread. Finding a general methodology for verifying such programs is challenging due to both the number of different tools involved and the intricate interface between the ``neural'' and ``symbolic'' program components. In this paper we present a general decomposition of the neuro-symbolic verification problem into parts, and examine the problem of the embedding gap that occurs when one tries to combine proofs about the neural and symbolic components. To address this problem we then introduce Vehicle -- standing as an abbreviation for a ``verification condition language'' -- an intermediate programming language interface between machine learning frameworks, automated theorem provers, and dependently-typed formalisations of neuro-symbolic programs. Vehicle allows users to specify the properties of the neural components of neuro-symbolic programs once, and then safely compile the specification to each interface using a tailored typing and compilation procedure. We give a high-level overview of Vehicle's overall design, its interfaces and compilation & type-checking procedures, and then demonstrate its utility by formally verifying the safety of a simple autonomous car controlled by a neural network, operating in a stochastic environment with imperfect information.
Authors: Yuquan Xie, Shengtao Peng, Wanqi Yang, Ming Yang, Yang Gao
Abstract: Knowledge Tracing (KT) is a critical task in online education systems, aiming to monitor students' knowledge states throughout a learning period. Common KT approaches involve predicting the probability of a student correctly answering the next question based on their exercise history. However, these methods often suffer from performance degradation when faced with the scarcity of student interactions in new education systems. To address this, we leverage student interactions from existing education systems to mitigate performance degradation caused by limited training data. Nevertheless, these interactions exhibit significant differences since they are derived from different education systems. To address this issue, we propose a domain generalization approach for knowledge tracing, where existing education systems are considered source domains, and new education systems with limited data are considered target domains. Additionally, we design a domain-generalizable knowledge tracing framework (DGKT) that can be applied to any KT model. Specifically, we present a concept aggregation approach designed to reduce conceptual disparities within sequences of student interactions from diverse domains. To further mitigate domain discrepancies, we introduce a novel normalization module called Sequence Instance Normalization (SeqIN). Moreover, to fully leverage exercise information, we propose a new knowledge tracing model tailored for the domain generalization KT task, named Domain-Generalizable Relation-based Knowledge Tracing (DGRKT). Extensive experiments across five benchmark datasets demonstrate that the proposed method performs well despite limited training data.
Authors: Zhaotian Weng, Zijun Gao, Jerone Andrews, Jieyu Zhao
Abstract: Vision-language models (VLMs) pre-trained on extensive datasets can inadvertently learn biases by correlating gender information with specific objects or scenarios. Current methods, which focus on modifying inputs and monitoring changes in the model's output probability scores, often struggle to comprehensively understand bias from the perspective of model components. We propose a framework that incorporates causal mediation analysis to measure and map the pathways of bias generation and propagation within VLMs. This approach allows us to identify the direct effects of interventions on model bias and the indirect effects of interventions on bias mediated through different model components. Our results show that image features are the primary contributors to bias, with significantly higher impacts than text features, specifically accounting for 32.57% and 12.63% of the bias in the MSCOCO and PASCAL-SENTENCE datasets, respectively. Notably, the image encoder's contribution surpasses that of the text encoder and the deep fusion encoder. Further experimentation confirms that contributions from both language and vision modalities are aligned and non-conflicting. Consequently, focusing on blurring gender representations within the image encoder, which contributes most to the model bias, reduces bias efficiently by 22.03% and 9.04% in the MSCOCO and PASCAL-SENTENCE datasets, respectively, with minimal performance loss or increased computational demands.
Authors: Zirui Shao, Feiyu Gao, Zhaoqing Zhu, Chuwei Luo, Hangdi Xing, Zhi Yu, Qi Zheng, Ming Yan, Jiajun Bu
Abstract: Multimodal large language models (MLLMs) have shown impressive capabilities in document understanding, a rapidly growing research area with significant industrial demand. As a multimodal task, document understanding requires models to possess both perceptual and cognitive abilities. However, due to different types of annotation noise in training, current MLLMs often face conflicts between perception and cognition. Taking a document VQA task (cognition) as an example, an MLLM might generate answers that do not match the corresponding visual content identified by its OCR (perception). This conflict suggests that the MLLM might struggle to establish an intrinsic connection between the information it "sees" and what it "understands". Such conflicts challenge the intuitive notion that cognition is consistent with perception, hindering the performance and explainability of MLLMs. In this paper, we define the conflicts between cognition and perception as Cognition and Perception (C&P) knowledge conflicts, a form of multimodal knowledge conflict, and systematically assess them with a focus on document understanding. Our analysis reveals that even GPT-4o, a leading MLLM, achieves only 75.26% C&P consistency. To mitigate the C&P knowledge conflicts, we propose a novel method called Multimodal Knowledge Consistency Fine-tuning. Our method reduces C&P knowledge conflicts across all tested MLLMs and enhances their performance in both cognitive and perceptual tasks. All data we construct will be publicly available.
Authors: Rupak Majumdar, Mahmoud Salamati, Sadegh Soudjani
Abstract: Learning to control an unknown dynamical system with respect to high-level temporal specifications is an important problem in control theory. We present the first regret-free online algorithm for learning a controller for linear temporal logic (LTL) specifications for systems with unknown dynamics. We assume that the underlying (unknown) dynamics is modeled by a finite-state and action Markov decision process (MDP). Our core technical result is a regret-free learning algorithm for infinite-horizon reach-avoid problems on MDPs. For general LTL specifications, we show that the synthesis problem can be reduced to a reach-avoid problem once the graph structure is known. Additionally, we provide an algorithm for learning the graph structure, assuming knowledge of a minimum transition probability, which operates independently of the main regret-free algorithm. Our LTL controller synthesis algorithm provides sharp bounds on how close we are to achieving optimal behavior after a finite number of learning episodes. In contrast, previous algorithms for LTL synthesis only provide asymptotic guarantees, which give no insight into the transient performance during the learning phase.
Authors: Shuxiang Cao, Zijian Zhang, Mohammed Alghadeer, Simone D Fasciati, Michele Piscitelli, Mustafa Bakr, Peter Leek, Al\'an Aspuru-Guzik
Abstract: Fully automated self-driving laboratories are promising to enable high-throughput and large-scale scientific discovery by reducing repetitive labour. However, effective automation requires deep integration of laboratory knowledge, which is often unstructured, multimodal, and difficult to incorporate into current AI systems. This paper introduces the k-agents framework, designed to support experimentalists in organizing laboratory knowledge and automating experiments with agents. Our framework employs large language model-based agents to encapsulate laboratory knowledge including available laboratory operations and methods for analyzing experiment results. To automate experiments, we introduce execution agents that break multi-step experimental procedures into agent-based state machines, interact with other agents to execute each step and analyze the experiment results. The analyzed results are then utilized to drive state transitions, enabling closed-loop feedback control. To demonstrate its capabilities, we applied the agents to calibrate and operate a superconducting quantum processor, where they autonomously planned and executed experiments for hours, successfully producing and characterizing entangled quantum states at the level achieved by human scientists. Our knowledge-based agent system opens up new possibilities for managing laboratory knowledge and accelerating scientific discovery.
Authors: Matthew Riemer, Zahra Ashktorab, Djallel Bouneffouf, Payel Das, Miao Liu, Justin D. Weisz, Murray Campbell
Abstract: Our paper argues that the majority of theory of mind benchmarks are broken because of their inability to directly test how large language models (LLMs) adapt to new partners. This problem stems from the fact that theory of mind benchmarks for LLMs are overwhelmingly inspired by the methods used to test theory of mind in humans and fall victim to a fallacy of attributing human-like qualities to AI agents. We expect that humans will engage in a consistent reasoning process across various questions about a situation, but this is known to not be the case for current LLMs. Most theory of mind benchmarks only measure what we call literal theory of mind: the ability to predict the behavior of others. However, this type of metric is only informative when agents exhibit self-consistent reasoning. Thus, we introduce the concept of functional theory of mind: the ability to adapt to agents in-context following a rational response to their behavior. We find that many open source LLMs are capable of displaying strong literal theory of mind capabilities, but seem to struggle with functional theory of mind -- even with exceedingly simple partner policies. Simply put, strong literal theory of mind performance does not necessarily imply strong functional theory of mind performance or vice versa. Achieving functional theory of mind, particularly over long interaction horizons with a partner, is a significant challenge deserving a prominent role in any meaningful LLM theory of mind evaluation.
Authors: Nantheera Anantrasirichai, Fan Zhang, David Bull
Abstract: The rapid advancements in artificial intelligence (AI), particularly in generative AI and large language models (LLMs), have profoundly impacted the creative industries, enabling more innovative content creation, enhancing workflows, and democratizing access to creative tools. This paper explores these technological shifts, with particular focus on how those that have emerged since our previous review in 2022 have expanded creative opportunities and improved efficiency. These technological advancements have enhanced the capabilities of text-to-image, text-to-video, and multimodal generation technologies. In particular, key breakthroughs in LLMs have established new benchmarks in conversational AI, while advancements in image generators have revolutionized content creation. We also discuss the integration of AI into post-production workflows, which has significantly accelerated and improved traditional processes. Once content has been created, it must be delivered to its audiences the media industry is facing the demands of increased communication traffic due to creative content. We therefore include a discussion of how AI is beginning to transform the way we represent and compress media content. We highlight the trend toward unified AI frameworks capable of addressing and integrating multiple creative tasks, and we underscore the importance of human insight to drive the creative process and oversight to mitigate AI-generated inaccuracies. Finally, we explore AI's future potential in the creative sector, stressing the need to navigate emerging challenges and to maximize its benefits while addressing the associated risks.
Authors: Nicole C. Wang
Abstract: This exploratory study investigates the intersection of Generative AI tools and experiential learning in business education. Through a case study of an innovative undergraduate course, we examine how students interact with and adapt to various AI modalities-from text-based tools to image generation-alongside real-world experiences. Our findings reveal how this integrated approach enables novice users to overcome creative barriers, accelerates skill acquisition, and creates a dynamic interplay between AI-generated insights and real-world validation. We identify critical interaction challenges, including prompt engineering patterns and the need for more intuitive AI interfaces in educational contexts. These insights inform the design of future AI tools for creative learning and contribute to broader HCI discussions about human-AI collaboration in educational settings.
Authors: Yuxin Zuo, Shang Qu, Yifei Li, Zhangren Chen, Xuekai Zhu, Ermo Hua, Kaiyan Zhang, Ning Ding, Bowen Zhou
Abstract: We introduce MedXpertQA, a highly challenging and comprehensive benchmark to evaluate expert-level medical knowledge and advanced reasoning. MedXpertQA includes 4,460 questions spanning 17 specialties and 11 body systems. It includes two subsets, Text for text evaluation and MM for multimodal evaluation. Notably, MM introduces expert-level exam questions with diverse images and rich clinical information, including patient records and examination results, setting it apart from traditional medical multimodal benchmarks with simple QA pairs generated from image captions. MedXpertQA applies rigorous filtering and augmentation to address the insufficient difficulty of existing benchmarks like MedQA, and incorporates specialty board questions to improve clinical relevance and comprehensiveness. We perform data synthesis to mitigate data leakage risk and conduct multiple rounds of expert reviews to ensure accuracy and reliability. We evaluate 18 leading models on \benchmark. Moreover, medicine is deeply connected to real-world decision-making, providing a rich and representative setting for assessing reasoning abilities beyond mathematics and code. To this end, we develop a reasoning-oriented subset to facilitate the assessment of o1-like models. Code and data are available at: https://github.com/TsinghuaC3I/MedXpertQA
Authors: Jiahe Jin, Yanheng He, Mingyan Yang
Abstract: In this work, we identify the "2D-Cheating" problem in 3D LLM evaluation, where these tasks might be easily solved by VLMs with rendered images of point clouds, exposing ineffective evaluation of 3D LLMs' unique 3D capabilities. We test VLM performance across multiple 3D LLM benchmarks and, using this as a reference, propose principles for better assessing genuine 3D understanding. We also advocate explicitly separating 3D abilities from 1D or 2D aspects when evaluating 3D LLMs. Code and data are available at https://github.com/LLM-class-group/Revisiting-3D-LLM-Benchmarks
URLs: https://github.com/LLM-class-group/Revisiting-3D-LLM-Benchmarks
Authors: Hanna M. Dettki, Brenden M. Lake, Charley M. Wu, Bob Rehder
Abstract: Causal reasoning is a core component of intelligence. Large language models (LLMs) have shown impressive capabilities in generating human-like text, raising questions about whether their responses reflect true understanding or statistical patterns. We compared causal reasoning in humans and four LLMs using tasks based on collider graphs, rating the likelihood of a query variable occurring given evidence from other variables. LLMs' causal inferences ranged from often nonsensical (GPT-3.5) to human-like to often more normatively aligned than those of humans (GPT-4o, Gemini-Pro, and Claude). Computational model fitting showed that one reason for GPT-4o, Gemini-Pro, and Claude's superior performance is they didn't exhibit the "associative bias" that plagues human causal reasoning. Nevertheless, even these LLMs did not fully capture subtler reasoning patterns associated with collider graphs, such as "explaining away".
Authors: Yi Xu, Laura Ruis, Tim Rockt\"aschel, Robert Kirk
Abstract: Automatic evaluation methods based on large language models (LLMs) are emerging as the standard tool for assessing the instruction-following abilities of LLM-based agents. The most common method in this paradigm, pairwise comparisons with a baseline model, critically depends on the assumption of transitive preferences. However, the validity of this assumption remains largely unexplored. In this study, we investigate the presence of non-transitivity within the AlpacaEval framework and analyze its effects on model rankings. We find that LLM judges exhibit non-transitive preferences, leading to rankings that are sensitive to the choice of the baseline model. To mitigate this issue, we show that round-robin tournaments combined with Bradley-Terry models of preference can produce more reliable rankings. Notably, our method increases both the Spearman correlation and the Kendall correlation with Chatbot Arena (95.0% -> 96.4% and 82.1% -> 86.3% respectively). To address the computational cost of round-robin tournaments, we propose Swiss-Wise Iterative Matchmaking (Swim) tournaments, using a dynamic matching strategy to capture the benefits of round-robin tournaments while maintaining computational efficiency.
Authors: John Burden, Marko Te\v{s}i\'c, Lorenzo Pacchiardi, Jos\'e Hern\'andez-Orallo
Abstract: Research in AI evaluation has grown increasingly complex and multidisciplinary, attracting researchers with diverse backgrounds and objectives. As a result, divergent evaluation paradigms have emerged, often developing in isolation, adopting conflicting terminologies, and overlooking each other's contributions. This fragmentation has led to insular research trajectories and communication barriers both among different paradigms and with the general public, contributing to unmet expectations for deployed AI systems. To help bridge this insularity, in this paper we survey recent work in the AI evaluation landscape and identify six main paradigms. We characterise major recent contributions within each paradigm across key dimensions related to their goals, methodologies and research cultures. By clarifying the unique combination of questions and approaches associated with each paradigm, we aim to increase awareness of the breadth of current evaluation approaches and foster cross-pollination between different paradigms. We also identify potential gaps in the field to inspire future research directions.
Authors: Kyle Gao, Dening Lu, Liangzhi Li, Nan Chen, Hongjie He, Linlin Xu, Jonathan Li
Abstract: The Los Angeles wildfires of January 2025 caused more than 250 billion dollars in damage and lasted for nearly an entire month before containment. Following our previous work, the Digital Twin Building, we modify and leverage the multi-agent large language model framework as well as the cloud-mapping integration to study the air quality during the Los Angeles wildfires. Recent advances in large language models have allowed for out-of-the-box automated large-scale data analysis. We use a multi-agent large language system comprised of an Instructor agent and Worker agents. Upon receiving the users' instructions, the Instructor agent retrieves the data from the cloud platform and produces instruction prompts to the Worker agents. The Worker agents then analyze the data and provide summaries. The summaries are finally input back into the Instructor agent, which then provides the final data analysis. We test this system's capability for data-based policy recommendation by assessing our Instructor-Worker LLM system's health recommendations based on air quality during the Los Angeles wildfires.
Authors: Jinwei Su, Yinghui Xia, Ronghua Shi, Jianhui Wang, Jianuo Huang, Yijin Wang, Tianyu Shi, Yang Jingsong, Lewei He
Abstract: Large language models (LLMs) have demonstrated strong potential and impressive performance in automating the generation and optimization of workflows. However, existing approaches are marked by limited reasoning capabilities, high computational demands, and significant resource requirements. To address these issues, we propose DebFlow, a framework that employs a debate mechanism to optimize workflows and integrates reflexion to improve based on previous experiences. We evaluated our method across six benchmark datasets, including HotpotQA, MATH, and ALFWorld. Our approach achieved a 3\% average performance improvement over the latest baselines, demonstrating its effectiveness in diverse problem domains. In particular, during training, our framework reduces resource consumption by 37\% compared to the state-of-the-art baselines. Additionally, we performed ablation studies. Removing the Debate component resulted in a 4\% performance drop across two benchmark datasets, significantly greater than the 2\% drop observed when the Reflection component was removed. These findings strongly demonstrate the critical role of Debate in enhancing framework performance, while also highlighting the auxiliary contribution of reflexion to overall optimization.
Authors: Xiangqi Wang, Yue Huang, Yanbo Wang, Xiaonan Luo, Kehan Guo, Yujun Zhou, Xiangliang Zhang
Abstract: LLMs often need effective configurations, like temperature and reasoning steps, to handle tasks requiring sophisticated reasoning and problem-solving, ranging from joke generation to mathematical reasoning. Existing prompting approaches usually adopt general-purpose, fixed configurations that work 'well enough' across tasks but seldom achieve task-specific optimality. To address this gap, we introduce AdaReasoner, an LLM-agnostic plugin designed for any LLM to automate adaptive reasoning configurations for tasks requiring different types of thinking. AdaReasoner is trained using a reinforcement learning (RL) framework, combining a factorized action space with a targeted exploration strategy, along with a pretrained reward model to optimize the policy model for reasoning configurations with only a few-shot guide. AdaReasoner is backed by theoretical guarantees and experiments of fast convergence and a sublinear policy gap. Across six different LLMs and a variety of reasoning tasks, it consistently outperforms standard baselines, preserves out-of-distribution robustness, and yield gains on knowledge-intensive tasks through tailored prompts.
Authors: Zeyu Liu, Yuhang Liu, Guanghao Zhu, Congkai Xie, Zhen Li, Jianbo Yuan, Xinyao Wang, Qing Li, Shing-Chi Cheung, Shengyu Zhang, Fei Wu, Hongxia Yang
Abstract: Recent advancements in large language models (LLMs) have demonstrated substantial progress in reasoning capabilities, such as DeepSeek-R1, which leverages rule-based reinforcement learning to enhance logical reasoning significantly. However, extending these achievements to multimodal large language models (MLLMs) presents critical challenges, which are frequently more pronounced for Multimodal Small Language Models (MSLMs) given their typically weaker foundational reasoning abilities: (1) the scarcity of high-quality multimodal reasoning datasets, (2) the degradation of reasoning capabilities due to the integration of visual processing, and (3) the risk that direct application of reinforcement learning may produce complex yet incorrect reasoning processes. To address these challenges, we design a novel framework Infi-MMR to systematically unlock the reasoning potential of MSLMs through a curriculum of three carefully structured phases and propose our multimodal reasoning model Infi-MMR-3B. The first phase, Foundational Reasoning Activation, leverages high-quality textual reasoning datasets to activate and strengthen the model's logical reasoning capabilities. The second phase, Cross-Modal Reasoning Adaptation, utilizes caption-augmented multimodal data to facilitate the progressive transfer of reasoning skills to multimodal contexts. The third phase, Multimodal Reasoning Enhancement, employs curated, caption-free multimodal data to mitigate linguistic biases and promote robust cross-modal reasoning. Infi-MMR-3B achieves both state-of-the-art multimodal math reasoning ability (43.68% on MathVerse testmini, 27.04% on MathVision test, and 21.33% on OlympiadBench) and general reasoning ability (67.2% on MathVista testmini). Resources are available at https://huggingface.co/Reallm-Labs/Infi-MMR-3B.
Authors: Yibo Zhao, Jiapeng Zhu, Ye Guo, Kangkang He, Xiang Li
Abstract: Graph-based RAG methods like GraphRAG have shown promising global understanding of the knowledge base by constructing hierarchical entity graphs. However, they often suffer from inefficiency and rely on manually pre-defined query modes, limiting practical use. In this paper, we propose E^2GraphRAG, a streamlined graph-based RAG framework that improves both Efficiency and Effectiveness. During the indexing stage, E^2GraphRAG constructs a summary tree with large language models and an entity graph with SpaCy based on document chunks. We then construct bidirectional indexes between entities and chunks to capture their many-to-many relationships, enabling fast lookup during both local and global retrieval. For the retrieval stage, we design an adaptive retrieval strategy that leverages the graph structure to retrieve and select between local and global modes. Experiments show that E^2GraphRAG achieves up to 10 times faster indexing than GraphRAG and 100 times speedup over LightRAG in retrieval while maintaining competitive QA performance.
Authors: Florian Carichon, Aditi Khandelwal, Marylou Fauchard, Golnoosh Farnadi
Abstract: This position paper states that AI Alignment in Multi-Agent Systems (MAS) should be considered a dynamic and interaction-dependent process that heavily depends on the social environment where agents are deployed, either collaborative, cooperative, or competitive. While AI alignment with human values and preferences remains a core challenge, the growing prevalence of MAS in real-world applications introduces a new dynamic that reshapes how agents pursue goals and interact to accomplish various tasks. As agents engage with one another, they must coordinate to accomplish both individual and collective goals. However, this complex social organization may unintentionally misalign some or all of these agents with human values or user preferences. Drawing on social sciences, we analyze how social structure can deter or shatter group and individual values. Based on these analyses, we call on the AI community to treat human, preferential, and objective alignment as an interdependent concept, rather than isolated problems. Finally, we emphasize the urgent need for simulation environments, benchmarks, and evaluation frameworks that allow researchers to assess alignment in these interactive multi-agent contexts before such dynamics grow too complex to control.
Authors: Renyang Liu, Wenjie Feng, Tianwei Zhang, Wei Zhou, Xueqi Cheng, See-Kiong Ng
Abstract: With the surge and widespread application of image generation models, data privacy and content safety have become major concerns and attracted great attention from users, service providers, and policymakers. Machine unlearning (MU) is recognized as a cost-effective and promising means to address these challenges. Despite some advancements, image generation model unlearning (IGMU) still faces remarkable gaps in practice, e.g., unclear task discrimination and unlearning guidelines, lack of an effective evaluation framework, and unreliable evaluation metrics. These can hinder the understanding of unlearning mechanisms and the design of practical unlearning algorithms. We perform exhaustive assessments over existing state-of-the-art unlearning algorithms and evaluation standards, and discover several critical flaws and challenges in IGMU tasks. Driven by these limitations, we make several core contributions, to facilitate the comprehensive understanding, standardized categorization, and reliable evaluation of IGMU. Specifically, (1) We design CatIGMU, a novel hierarchical task categorization framework. It provides detailed implementation guidance for IGMU, assisting in the design of unlearning algorithms and the construction of testbeds. (2) We introduce EvalIGMU, a comprehensive evaluation framework. It includes reliable quantitative metrics across five critical aspects. (3) We construct DataIGM, a high-quality unlearning dataset, which can be used for extensive evaluations of IGMU, training content detectors for judgment, and benchmarking the state-of-the-art unlearning algorithms. With EvalIGMU and DataIGM, we discover that most existing IGMU algorithms cannot handle the unlearning well across different evaluation dimensions, especially for preservation and robustness. Code and models are available at https://github.com/ryliu68/IGMU.
Authors: Violet Xiang, Chase Blagden, Rafael Rafailov, Nathan Lile, Sang Truong, Chelsea Finn, Nick Haber
Abstract: Large reasoning models (LRMs) achieve higher performance on challenging reasoning tasks by generating more tokens at inference time, but this verbosity often wastes computation on easy problems. Existing solutions, including supervised finetuning on shorter traces, user-controlled budgets, or RL with uniform penalties, either require data curation, manual configuration, or treat all problems alike regardless of difficulty. We introduce Adaptive Length Penalty (ALP), a reinforcement learning objective tailoring generation length to per-prompt solve rate. During training, ALP monitors each prompt's online solve rate through multiple rollouts and adds a differentiable penalty whose magnitude scales inversely with that rate, so confident (easy) prompts incur a high cost for extra tokens while hard prompts remain unhindered. Posttraining DeepScaleR-1.5B with ALP cuts average token usage by 50\% without significantly dropping performance. Relative to fixed-budget and uniform penalty baselines, ALP redistributes its reduced budget more intelligently by cutting compute on easy prompts and reallocating saved tokens to difficult ones, delivering higher accuracy on the hardest problems with higher cost.
Authors: Suzan Ece Ada, Erhan Oztop, Emre Ugur
Abstract: Offline Reinforcement Learning (RL) methods leverage previous experiences to learn better policies than the behavior policy used for data collection. However, they face challenges handling distribution shifts due to the lack of online interaction during training. To this end, we propose a novel method named State Reconstruction for Diffusion Policies (SRDP) that incorporates state reconstruction feature learning in the recent class of diffusion policies to address the problem of out-of-distribution (OOD) generalization. Our method promotes learning of generalizable state representation to alleviate the distribution shift caused by OOD states. To illustrate the OOD generalization and faster convergence of SRDP, we design a novel 2D Multimodal Contextual Bandit environment and realize it on a 6-DoF real-world UR10 robot, as well as in simulation, and compare its performance with prior algorithms. In particular, we show the importance of the proposed state reconstruction via ablation studies. In addition, we assess the performance of our model on standard continuous control benchmarks (D4RL), namely the navigation of an 8-DoF ant and forward locomotion of half-cheetah, hopper, and walker2d, achieving state-of-the-art results. Finally, we demonstrate that our method can achieve 167% improvement over the competing baseline on a sparse continuous control navigation task where various regions of the state space are removed from the offline RL dataset, including the region encapsulating the goal.
Authors: Andrea Cini, Ivan Marisca, Daniele Zambon, Cesare Alippi
Abstract: Graph deep learning methods have become popular tools to process collections of correlated time series. Unlike traditional multivariate forecasting methods, graph-based predictors leverage pairwise relationships by conditioning forecasts on graphs spanning the time series collection. The conditioning takes the form of architectural inductive biases on the forecasting architecture, resulting in a family of models called spatiotemporal graph neural networks. These biases allow for training global forecasting models on large collections of time series while localizing predictions w.r.t. each element in the set (nodes) by accounting for correlations among them (edges). Recent advances in graph neural networks and deep learning for time series forecasting make the adoption of such processing framework appealing and timely. However, most studies focus on refining existing architectures by exploiting modern deep-learning practices. Conversely, foundational and methodological aspects have not been subject to systematic investigation. To fill this void, this tutorial paper aims to introduce a comprehensive methodological framework formalizing the forecasting problem and providing design principles for graph-based predictors, as well as methods to assess their performance. In addition, together with an overview of the field, we provide design guidelines and best practices, as well as an in-depth discussion of open challenges and future directions.
Authors: Qifang Zhao, Weidong Ren, Tianyu Li, Hong Liu, Xingsheng He, Xiaoxiao Xu
Abstract: We introduceGraphGPT, a novel self-supervised generative pre-trained model for graph learning based on the Graph Eulerian Transformer (GET). First, we propose GET, which combines a standard transformer encoder or decoder architecture with an innovative graph-to-sequence transformation method. This method converts graphs or sampled subgraphs into sequences of tokens representing nodes, edges, and attributes in a reversible manner using Eulerian paths. We pre-train GET using either of the two self-supervised tasks: next-token prediction (NTP) and scheduled masked-token prediction (SMTP). The pre-trained model is then fine-tuned for downstream tasks such as graph-, edge-, and node-level prediction. Despite its simplicity, GraphGPT achieves performance comparable to or surpassing state-of-the-art methods on multiple large-scale Open Graph Benchmark (OGB) datasets. It demonstrates exceptional results on the molecular property prediction dataset PCQM4Mv2 and the protein-protein interaction dataset ogbl-ppa. Notably, generative pre-training enables scaling GraphGPT to 2 billion parameters while maintaining performance gains - a breakthrough that overcomes the scalability limitations of traditional Graph Neural Networks (GNNs) and prior graph transformers (GTs). To advance research in graph foundation models and facilitate scientific discovery in chemistry, materials science, and related fields, we will release the source code (https://github.com/alibaba/graph-gpt) and pre-trained checkpoints.
Authors: Qinggang Zhang, Hao Chen, Junnan Dong, Shengyuan Chen, Feiran Huang, Xiao Huang
Abstract: Recent advancements in large language models (LLMs) have shown promise in bridging the gap between natural language queries and database management systems, enabling users to interact with databases without the background of SQL. However, LLMs often struggle to comprehend complex database structures and accurately interpret user intentions. Decomposition-based methods have been proposed to enhance the performance of LLMs on complex tasks, but decomposing SQL generation into subtasks is non-trivial due to the declarative structure of SQL syntax and the intricate connections between query concepts and database elements. In this paper, we propose a novel Structure GUided text-to-SQL framework~(SGU-SQL) that incorporates syntax-based prompting to enhance the SQL generation capabilities of LLMs. Specifically, SGU-SQL establishes structure-aware links between user queries and database schema and decomposes the complex generation task using syntax-based prompting to enable more accurate LLM-based SQL generation. Extensive experiments on two benchmark datasets demonstrate that SGU-SQL consistently outperforms state-of-the-art text-to-SQL models.
Authors: Minsu Kim, Jee-weon Jung, Hyeongseop Rha, Soumi Maiti, Siddhant Arora, Xuankai Chang, Shinji Watanabe, Yong Man Ro
Abstract: The capability to jointly process multi-modal information is becoming an essential task. However, the limited number of paired multi-modal data and the large computational requirements in multi-modal learning hinder the development. We propose a novel Tri-Modal Translation (TMT) model that translates between arbitrary modalities spanning speech, image, and text. We introduce a novel viewpoint, where we interpret different modalities as different languages, and treat multi-modal translation as a well-established machine translation problem. To this end, we tokenize speech and image data into discrete tokens, which provide a unified interface across modalities and significantly decrease the computational cost. In the proposed TMT, a multi-modal encoder-decoder conducts the core translation, whereas modality-specific processing is conducted only within the tokenization and detokenization stages. We evaluate the proposed TMT on all six modality translation tasks. TMT outperforms single model counterparts consistently, demonstrating that unifying tasks is beneficial not only for practicality but also for performance.
Authors: Toru Shirakawa, Yi Li, Yulun Wu, Sky Qiu, Yuxuan Li, Mingduo Zhao, Hiroyasu Iso, Mark van der Laan
Abstract: We propose Deep Longitudinal Targeted Minimum Loss-based Estimation (Deep LTMLE), a novel approach to estimate the counterfactual mean of outcome under dynamic treatment policies in longitudinal problem settings. Our approach utilizes a transformer architecture with heterogeneous type embedding trained using temporal-difference learning. After obtaining an initial estimate using the transformer, following the targeted minimum loss-based likelihood estimation (TMLE) framework, we statistically corrected for the bias commonly associated with machine learning algorithms. Furthermore, our method also facilitates statistical inference by enabling the provision of 95% confidence intervals grounded in asymptotic statistical theory. Simulation results demonstrate our method's superior performance over existing approaches, particularly in complex, long time-horizon scenarios. It remains effective in small-sample, short-duration contexts, matching the performance of asymptotically efficient estimators. To demonstrate our method in practice, we applied our method to estimate counterfactual mean outcomes for standard versus intensive blood pressure management strategies in a real-world cardiovascular epidemiology cohort study.
Authors: Dohoon Lee, Jaehyun Park, Hyunwoo J. Kim, Kyogu Lee
Abstract: Flow and diffusion models have demonstrated strong performance and training stability across various tasks but lack two critical properties of simulation-based methods: freedom of dimensionality and adaptability to different inference trajectories. To address this limitation, we propose the Multidimensional Adaptive Coefficient (MAC), a plug-in module for flow and diffusion models that extends conventional unidimensional coefficients to multidimensional ones and enables inference trajectory-wise adaptation. MAC is trained via simulation-based feedback through adversarial refinement. Empirical results across diverse frameworks and datasets demonstrate that MAC enhances generative quality with high training efficiency. Consequently, our work offers a new perspective on inference trajectory optimality, encouraging future research to move beyond vector field design and to leverage training-efficient, simulation-based optimization.
Authors: Mengdi Wu, Xinhao Cheng, Shengyu Liu, Chunan Shi, Jianan Ji, Kit Ao, Praveen Velliengiri, Xupeng Miao, Oded Padon, Zhihao Jia
Abstract: We introduce Mirage, the first multi-level superoptimizer for tensor programs. A key idea in Mirage is $\mu$Graphs, a uniform representation of tensor programs at the kernel, thread block, and thread levels of the GPU compute hierarchy. $\mu$Graphs enable Mirage to discover novel optimizations that combine algebraic transformations, schedule transformations, and generation of new custom kernels. To navigate the large search space, Mirage introduces a pruning technique based on abstraction that significantly reduces the search space and provides a certain optimality guarantee. To ensure that the optimized $\mu$Graph is equivalent to the input program, Mirage introduces a probabilistic equivalence verification procedure with strong theoretical guarantees. Our evaluation shows that Mirage outperforms existing approaches by up to 3.3$\times$ even for DNNs that are widely used and heavily optimized. Mirage is publicly available at https://github.com/mirage-project/mirage.
Authors: Lun Ai, Stephen H. Muggleton, Shi-Shun Liang, Geoff S. Baldwin
Abstract: Reasoning about hypotheses and updating knowledge through empirical observations are central to scientific discovery. In this work, we applied logic-based machine learning methods to drive biological discovery by guiding experimentation. Genome-scale metabolic network models (GEMs) - comprehensive representations of metabolic genes and reactions - are widely used to evaluate genetic engineering of biological systems. However, GEMs often fail to accurately predict the behaviour of genetically engineered cells, primarily due to incomplete annotations of gene interactions. The task of learning the intricate genetic interactions within GEMs presents computational and empirical challenges. To efficiently predict using GEM, we describe a novel approach called Boolean Matrix Logic Programming (BMLP) by leveraging Boolean matrices to evaluate large logic programs. We developed a new system, $BMLP_{active}$, which guides cost-effective experimentation and uses interpretable logic programs to encode a state-of-the-art GEM of a model bacterial organism. Notably, $BMLP_{active}$ successfully learned the interaction between a gene pair with fewer training examples than random experimentation, overcoming the increase in experimental design space. $BMLP_{active}$ enables rapid optimisation of metabolic models to reliably engineer biological systems for producing useful compounds. It offers a realistic approach to creating a self-driving lab for biological discovery, which would then facilitate microbial engineering for practical applications.
Authors: Jerry Yao-Chieh Hu, Maojiang Su, En-Jui Kuo, Zhao Song, Han Liu
Abstract: We study the computational limits of Low-Rank Adaptation (LoRA) for finetuning transformer-based models using fine-grained complexity theory. Our key observation is that the existence of low-rank decompositions within the gradient computation of LoRA adaptation leads to possible algorithmic speedup. This allows us to (i) identify a phase transition behavior of efficiency assuming the Strong Exponential Time Hypothesis (SETH), and (ii) prove the existence of almost linear algorithms by controlling the LoRA update computation term by term. For the former, we identify a sharp transition in the efficiency of all possible rank-$r$ LoRA update algorithms for transformers, based on specific norms resulting from the multiplications of the input sequence $X$, pretrained weights ${W^\star}$, and adapter matrices $\alpha B A/r$. Specifically, we derive a shared upper bound threshold for such norms, and show that efficient (sub-quadratic) approximation algorithms of LoRA exist only below this threshold. For the latter, we prove the existence of almost linear approximation algorithms for LoRA adaptation by utilizing the hierarchical low-rank structures of LoRA gradients and approximating the gradients with a series of chained low-rank approximations. To showcase our theory, we consider two practical scenarios: partial (e.g., only $W_V$ and $W_Q$) and full adaptations (e.g., $W_Q$, $W_V$, and $W_K$) of weights in attention heads.
Authors: Lukas Helff, Felix Friedrich, Manuel Brack, Kristian Kersting, Patrick Schramowski
Abstract: This paper introduces LlavaGuard, a suite of VLM-based vision safeguards that address the critical need for reliable guardrails in the era of large-scale data and models. To this end, we establish a novel open framework, describing a customizable safety taxonomy, data preprocessing, augmentation, and training setup. For teaching a VLM safeguard on safety, we further create a multimodal safety dataset with high-quality human expert annotations, where each image is labeled with a safety rating, category, and rationale. We also employ advanced augmentations to support context-specific assessments. The resulting LlavaGuard models, ranging from 0.5B to 7B, serve as a versatile tool for evaluating the safety compliance of visual content against flexible policies. In comprehensive experiments, LlavaGuard outperforms both state-of-the-art safeguards and VLMs in accuracy and in flexibly handling different policies. Additionally, we demonstrate LlavaGuard's performance in two real-world applications: large-scale dataset annotation and moderation of text-to-image models. We make our entire framework, including the dataset, model weights, and training code.
Authors: Zhuoyun Du, Chen Qian, Wei Liu, Zihao Xie, YiFei Wang, Rennai Qiu, Yufan Dang, Weize Chen, Cheng Yang, Ye Tian, Xuantang Xiong, Lei Han
Abstract: Large Language Models (LLMs) have significantly impacted various domains, especially through organized LLM-driven autonomous agents. A representative scenario is in software development, where agents can collaborate in a team like humans, following predefined phases to complete sub-tasks sequentially. However, for an agent team, each phase yields only one possible outcome. This results in the completion of only one development chain, thereby losing the opportunity to explore multiple potential decision paths within the solution space. Consequently leading to suboptimal results or extensive trial and error. To address this, we introduce Cross-Team Orchestration (Croto), a scalable multi-team framework that enables orchestrated teams to jointly propose various task-oriented solutions and interact with their insights in a self-independence while cross-team collaboration environment for superior solutions generation. Experiments reveal a notable increase in software quality compared to state-of-the-art baselines. We further tested our framework on story generation tasks, which demonstrated a promising generalization ability of our framework in other domains. The code and data is available at https://github.com/OpenBMB/ChatDev/tree/macnet
Authors: Matthew Wicker, Philip Sosnin, Igor Shilov, Adrianna Janik, Mark N. M\"uller, Yves-Alexandre de Montjoye, Adrian Weller, Calvin Tsay
Abstract: We study private prediction where differential privacy is achieved by adding noise to the outputs of a non-private model. Existing methods rely on noise proportional to the global sensitivity of the model, often resulting in sub-optimal privacy-utility trade-offs compared to private training. We introduce a novel approach for computing dataset-specific upper bounds on prediction sensitivity by leveraging convex relaxation and bound propagation techniques. By combining these bounds with the smooth sensitivity mechanism, we significantly improve the privacy analysis of private prediction compared to global sensitivity-based approaches. Experimental results across real-world datasets in medical image classification and natural language processing demonstrate that our sensitivity bounds are can be orders of magnitude tighter than global sensitivity. Our approach provides a strong basis for the development of novel privacy preserving technologies.
Authors: Junhan Kim, Ho-young Kim, Eulrang Cho, Chungman Lee, Joonyoung Kim, Yongkweon Jeon
Abstract: Post-training quantization (PTQ) is a promising solution for deploying large language models (LLMs) on resource-constrained devices. Early methods developed for small-scale networks, such as ResNet, rely on gradient-based optimization, which becomes impractical for hyper-scale LLMs with billions of parameters. While recently proposed backpropagation-free or transformation-based methods alleviate this issue, they ignore inter-layer interactions or use the naive nearest-rounding-based quantized weight assignment to save the heavy computational cost of weight optimization. In this paper, we introduce a novel backpropagation-free PTQ algorithm that optimizes quantized weights by considering inter-layer dependencies. The key innovation is the development of attention-aware Hessian matrices that capture inter-layer interactions within the attention module. Extensive experiments demonstrate that our approach not only outperforms existing weight quantization methods but also shows good synergy with conventional methods to suppress activation outliers, leading to state-of-the-art weight-activation quantization performance. The code will be available at https://github.com/SamsungLabs/BoA.
Authors: Giacomo Spigler
Abstract: We introduce Proximal Policy Distillation (PPD), a novel policy distillation method that integrates student-driven distillation and Proximal Policy Optimization (PPO) to increase sample efficiency and to leverage the additional rewards that the student policy collects during distillation. To assess the efficacy of our method, we compare PPD with two common alternatives, student-distill and teacher-distill, over a wide range of reinforcement learning environments that include discrete actions and continuous control (ATARI, Mujoco, and Procgen). For each environment and method, we perform distillation to a set of target student neural networks that are smaller, identical (self-distillation), or larger than the teacher network. Our findings indicate that PPD improves sample efficiency and produces better student policies compared to typical policy distillation approaches. Moreover, PPD demonstrates greater robustness than alternative methods when distilling policies from imperfect demonstrations. The code for the paper is released as part of a new Python library built on top of stable-baselines3 to facilitate policy distillation: `sb3-distill'.
Authors: Zifeng Ding, Yifeng Li, Yuan He, Antonio Norelli, Jingcheng Wu, Volker Tresp, Michael Bronstein, Yunpu Ma
Abstract: Learning useful representations for continuous-time dynamic graphs (CTDGs) is challenging, due to the concurrent need to span long node interaction histories and grasp nuanced temporal details. In particular, two problems emerge: (1) Encoding longer histories requires more computational resources, making it crucial for CTDG models to maintain low computational complexity to ensure efficiency; (2) Meanwhile, more powerful models are needed to identify and select the most critical temporal information within the extended context provided by longer histories. To address these problems, we propose a CTDG representation learning model named DyGMamba, originating from the popular Mamba state space model (SSM). DyGMamba first leverages a node-level SSM to encode the sequence of historical node interactions. Another time-level SSM is then employed to exploit the temporal patterns hidden in the historical graph, where its output is used to dynamically select the critical information from the interaction history. We validate DyGMamba experimentally on the dynamic link prediction task. The results show that our model achieves state-of-the-art in most cases. DyGMamba also maintains high efficiency in terms of computational resources, making it possible to capture long temporal dependencies with a limited computation budget.
Authors: Wentao Gao, Jiuyong Li, Debo Cheng, Lin Liu, Jixue Liu, Thuc Duy Le, Xiaojing Du, Xiongren Chen, Yanchang Zhao, Yun Chen
Abstract: Global Climate Models (GCMs) are crucial for predicting future climate changes by simulating the Earth systems. However, the GCM Outputs exhibit systematic biases due to model uncertainties, parameterization simplifications, and inadequate representation of complex climate phenomena. Traditional bias correction methods, which rely on historical observation data and statistical techniques, often neglect unobserved confounders, leading to biased results. This paper proposes a novel bias correction approach to utilize both GCM and observational data to learn a factor model that captures multi-cause latent confounders. Inspired by recent advances in causality based time series deconfounding, our method first constructs a factor model to learn latent confounders from historical data and then applies them to enhance the bias correction process using advanced time series forecasting models. The experimental results demonstrate significant improvements in the accuracy of precipitation outputs. By addressing unobserved confounders, our approach offers a robust and theoretically grounded solution for climate model bias correction.
Authors: Mouxiang Chen, Lefei Shen, Zhuo Li, Xiaoyun Joy Wang, Jianling Sun, Chenghao Liu
Abstract: Foundation models have emerged as a promising approach in time series forecasting (TSF). Existing approaches either repurpose large language models (LLMs) or build large-scale time series datasets to develop TSF foundation models for universal forecasting. However, these methods face challenges due to the severe cross-domain gap or in-domain heterogeneity. This paper explores a new road to building a TSF foundation model from rich, high-quality natural images. Our key insight is that a visual masked autoencoder, pre-trained on the ImageNet dataset, can naturally be a numeric series forecaster. By reformulating TSF as an image reconstruction task, we bridge the gap between image pre-training and TSF downstream tasks. Surprisingly, without further adaptation in the time series domain, the proposed VisionTS could achieve better zero-shot forecast performance than existing TSF foundation models. With fine-tuning for one epoch, VisionTS could further improve the forecasting and achieve state-of-the-art performance in most cases. Extensive experiments reveal intrinsic similarities between images and real-world time series, suggesting that visual models may offer a "free lunch" for TSF and highlight the potential for future cross-modality research. Our code is publicly available at https://github.com/Keytoyze/VisionTS.
Authors: Patara Trirat, Wonyong Jeong, Sung Ju Hwang
Abstract: Automated machine learning (AutoML) accelerates AI development by automating tasks in the development pipeline, such as optimal model search and hyperparameter tuning. Existing AutoML systems often require technical expertise to set up complex tools, which is in general time-consuming and requires a large amount of human effort. Therefore, recent works have started exploiting large language models (LLM) to lessen such burden and increase the usability of AutoML frameworks via a natural language interface, allowing non-expert users to build their data-driven solutions. These methods, however, are usually designed only for a particular process in the AI development pipeline and do not efficiently use the inherent capacity of the LLMs. This paper proposes AutoML-Agent, a novel multi-agent framework tailored for full-pipeline AutoML, i.e., from data retrieval to model deployment. AutoML-Agent takes user's task descriptions, facilitates collaboration between specialized LLM agents, and delivers deployment-ready models. Unlike existing work, instead of devising a single plan, we introduce a retrieval-augmented planning strategy to enhance exploration to search for more optimal plans. We also decompose each plan into sub-tasks (e.g., data preprocessing and neural network design) each of which is solved by a specialized agent we build via prompting executing in parallel, making the search process more efficient. Moreover, we propose a multi-stage verification to verify executed results and guide the code generation LLM in implementing successful solutions. Extensive experiments on seven downstream tasks using fourteen datasets show that AutoML-Agent achieves a higher success rate in automating the full AutoML process, yielding systems with good performance throughout the diverse domains.
Authors: Tingyu Zhu, Haoyu Liu, Ziyu Wang, Zhimin Jiang, Zeyu Zheng
Abstract: Developing generative models to create or conditionally create symbolic music presents unique challenges due to the combination of limited data availability and the need for high precision in note pitch. To address these challenges, we introduce an efficient Fine-Grained Guidance (FGG) approach within diffusion models. FGG guides the diffusion models to generate music that aligns more closely with the control and intent of expert composers, which is critical to improve the accuracy, listenability, and quality of generated music. This approach empowers diffusion models to excel in advanced applications such as improvisation, and interactive music creation. We derive theoretical characterizations for both the challenges in symbolic music generation and the effects of the FGG approach. We provide numerical experiments and subjective evaluation to demonstrate the effectiveness of our approach. We have published a demo page to showcase performances, which enables real-time interactive generation.
Authors: Tingchen Fu, Mrinank Sharma, Philip Torr, Shay B. Cohen, David Krueger, Fazl Barez
Abstract: Preference learning is a central component for aligning current LLMs, but this process can be vulnerable to data poisoning attacks. To address this concern, we introduce PoisonBench, a benchmark for evaluating large language models' susceptibility to data poisoning during preference learning. Data poisoning attacks can manipulate large language model responses to include hidden malicious content or biases, potentially causing the model to generate harmful or unintended outputs while appearing to function normally. We deploy two distinct attack types across eight realistic scenarios, assessing 21 widely-used models. Our findings reveal concerning trends: (1) Scaling up parameter size does not inherently enhance resilience against poisoning attacks; (2) There exists a log-linear relationship between the effects of the attack and the data poison ratio; (3) The effect of data poisoning can generalize to extrapolated triggers that are not included in the poisoned data. These results expose weaknesses in current preference learning techniques, highlighting the urgent need for more robust defenses against malicious models and data manipulation.
Authors: Irina Babayan, Hazhir Aliahmadi, Greg van Anders
Abstract: The broad range of neural network training techniques that invoke optimization but rely on ad hoc modification for validity suggests that optimization-based training is misguided. Shortcomings of optimization-based training are brought to particularly strong relief by the problem of overfitting, where naive optimization produces spurious outcomes. The broad success of neural networks for modelling physical processes has prompted advances that are based on inverting the direction of investigation and treating neural networks as if they were physical systems in their own right. These successes raise the question of whether broader, physical perspectives could motivate the construction of improved training algorithms. Here, we introduce simmering, a physics-based method that trains neural networks to generate weights and biases that are merely ``good enough'', but which, paradoxically, outperforms leading optimization-based approaches. Using classification and regression examples we show that simmering corrects neural networks that are overfit by Adam, and show that simmering avoids overfitting if deployed from the outset. Our results question optimization as a paradigm for neural network training, and leverage information-geometric arguments to point to the existence of classes of sufficient training algorithms that do not take optimization as their starting point.
Authors: Joongwon Chae, Zhenyu Wang, Ijaz Gul, Jiansong Ji, Zhenglin Chen, Peiwu Qin
Abstract: Recent advancements in protein structure prediction, particularly AlphaFold2, have revolutionized structural biology by achieving near-experimental accuracy ($\text{average RMSD} < 1.5\text{\AA}$). However, the computational demands of these models (approximately 30 minutes per protein on an RTX 4090) significantly limit their application in high-throughput protein screening. While large language models like ESM (Evolutionary Scale Modeling) have shown promise in extracting structural information directly from protein sequences, rapid assessment of protein structure quality for large-scale analyses remains a major challenge. We introduce pLDDT-Predictor, a high-speed protein screening tool that achieves a $250,000\times$ speedup compared to AlphaFold2 by leveraging pre-trained ESM2 protein embeddings and a Transformer architecture. Our model predicts AlphaFold2's pLDDT (predicted Local Distance Difference Test) scores with a Pearson correlation of 0.7891 and processes proteins in just 0.007 seconds on average. Using a comprehensive dataset of 1.5 million diverse protein sequences (ranging from 50 to 2048 amino acids), we demonstrate that pLDDT-Predictor accurately classifies high-confidence structures (pLDDT $>$ 70) with 91.2\% accuracy and achieves an MSE of 84.8142 compared to AlphaFold2's predictions. The source code and pre-trained models are freely available at https://github.com/jw-chae/pLDDT_Predictor, enabling the research community to perform rapid, large-scale protein structure quality assessments.
Authors: Elisabeth Kirsten, Ivan Habernal, Vedant Nanda, Muhammad Bilal Zafar
Abstract: Last few years have seen unprecedented advances in capabilities of Large Language Models (LLMs). These advancements promise to benefit a vast array of application domains. However, due to their immense size, performing inference with LLMs is both costly and slow. Consequently, a plethora of recent work has proposed strategies to enhance inference efficiency, e.g., quantization, pruning, and caching. These acceleration strategies reduce the inference cost and latency, often by several factors, while maintaining much of the predictive performance measured via common benchmarks. In this work, we explore another critical aspect of LLM performance: demographic bias in model generations due to inference acceleration optimizations. Using a wide range of metrics, we probe bias in model outputs from a number of angles. Analysis of outputs before and after inference acceleration shows significant change in bias. Worryingly, these bias effects are complex and unpredictable. A combination of an acceleration strategy and bias type may show little bias change in one model but may lead to a large effect in another. Our results highlight a need for in-depth and case-by-case evaluation of model bias after it has been modified to accelerate inference.
Authors: Tianyu Yang, Lisen Dai, Xiangqi Wang, Minhao Cheng, Yapeng Tian, Xiangliang Zhang
Abstract: Machine unlearning (MU) has gained significant attention as a means to remove specific data from trained models without requiring a full retraining process. While progress has been made in unimodal domains like text and image classification, unlearning in multimodal models remains relatively underexplored. In this work, we address the unique challenges of unlearning in CLIP, a prominent multimodal model that aligns visual and textual representations. We introduce CLIPErase, a novel approach that disentangles and selectively forgets both visual and textual associations, ensuring that unlearning does not compromise model performance. CLIPErase consists of three key modules: a Forgetting Module that disrupts the associations in the forget set, a Retention Module that preserves performance on the retain set, and a Consistency Module that maintains consistency with the original model. Extensive experiments on the CIFAR-100 and Flickr30K datasets across four CLIP downstream tasks demonstrate that CLIPErase effectively forgets designated associations in zero-shot tasks for multimodal samples, while preserving the model's performance on the retain set after unlearning.
Authors: Wanqi Yang, Yanda Li, Meng Fang, Yunchao Wei, Ling Chen
Abstract: Adversarial audio attacks pose a significant threat to the growing use of large audio-language models (LALMs) in voice-based human-machine interactions. While existing research focused on model-specific adversarial methods, real-world applications demand a more generalizable and universal approach to audio adversarial attacks. In this paper, we introduce the Chat-Audio Attacks (CAA) benchmark including four distinct types of audio attacks, which aims to explore the vulnerabilities of LALMs to these audio attacks in conversational scenarios. To evaluate the robustness of LALMs, we propose three evaluation strategies: Standard Evaluation, utilizing traditional metrics to quantify model performance under attacks; GPT-4o-Based Evaluation, which simulates real-world conversational complexities; and Human Evaluation, offering insights into user perception and trust. We evaluate six state-of-the-art LALMs with voice interaction capabilities, including Gemini-1.5-Pro, GPT-4o, and others, using three distinct evaluation methods on the CAA benchmark. Our comprehensive analysis reveals the impact of four types of audio attacks on the performance of these models, demonstrating that GPT-4o exhibits the highest level of resilience. Our data can be accessed via the following link: \href{https://github.com/crystraldo/CAA}{CAA}.
Authors: Jerry Yao-Chieh Hu, Wei-Po Wang, Ammar Gilani, Chenyang Li, Zhao Song, Han Liu
Abstract: We investigate the statistical and computational limits of prompt tuning for transformer-based foundation models. Our key contributions are prompt tuning on \emph{single-head} transformers with only a \emph{single} self-attention layer: (i) is universal, and (ii) supports efficient (even almost-linear time) algorithms under the Strong Exponential Time Hypothesis (SETH). Statistically, we prove that prompt tuning on such simplest possible transformers are universal approximators for sequence-to-sequence Lipschitz functions. In addition, we provide an exponential-in-$dL$ and -in-$(1/\epsilon)$ lower bound on the required soft-prompt tokens for prompt tuning to memorize any dataset with 1-layer, 1-head transformers. Computationally, we identify a phase transition in the efficiency of prompt tuning, determined by the norm of the \emph{soft-prompt-induced} keys and queries, and provide an upper bound criterion. Beyond this criterion, no sub-quadratic (efficient) algorithm for prompt tuning exists under SETH. Within this criterion, we showcase our theory by proving the existence of almost-linear time prompt tuning inference algorithms. These fundamental limits provide important necessary conditions for designing expressive and efficient prompt tuning methods for practitioners.
Authors: Varshita Kolipaka, Akshit Sinha, Debangan Mishra, Sumit Kumar, Arvindh Arun, Shashwat Goel, Ponnurangam Kumaraguru
Abstract: Graph Neural Networks (GNNs) are increasingly being used for a variety of ML applications on graph data. Because graph data does not follow the independently and identically distributed (i.i.d.) assumption, adversarial manipulations or incorrect data can propagate to other data points through message passing, which deteriorates the model's performance. To allow model developers to remove the adverse effects of manipulated entities from a trained GNN, we study the recently formulated problem of Corrective Unlearning. We find that current graph unlearning methods fail to unlearn the effect of manipulations even when the whole manipulated set is known. We introduce a new graph unlearning method, Cognac, which can unlearn the effect of the manipulation set even when only 5% of it is identified. It recovers most of the performance of a strong oracle with fully corrected training data, even beating retraining from scratch without the deletion set while being 8x more efficient. We hope our work assists GNN developers in mitigating harmful effects caused by issues in real-world data, post-training. Our code is publicly available at https://github.com/cognac-gnn-unlearning/corrective-unlearning-for-gnns
URLs: https://github.com/cognac-gnn-unlearning/corrective-unlearning-for-gnns
Authors: Dongjae Jeon, Dueun Kim, Albert No
Abstract: In this paper, we introduce a geometric framework to analyze memorization in diffusion models through the sharpness of the log probability density. We mathematically justify a previously proposed score-difference-based memorization metric by demonstrating its effectiveness in quantifying sharpness. Additionally, we propose a novel memorization metric that captures sharpness at the initial stage of image generation in latent diffusion models, offering early insights into potential memorization. Leveraging this metric, we develop a mitigation strategy that optimizes the initial noise of the generation process using a sharpness-aware regularization term.
Authors: Thomas Pouplin, Katarzyna Kobalczyk, Hao Sun, Mihaela van der Schaar
Abstract: Developing autonomous agents capable of performing complex, multi-step decision-making tasks specified in natural language remains a significant challenge, particularly in realistic settings where labeled data is scarce and real-time experimentation is impractical. Existing reinforcement learning (RL) approaches often struggle to generalize to unseen goals and states, limiting their applicability. In this paper, we introduce TEDUO, a novel training pipeline for offline language-conditioned policy learning in symbolic environments. Unlike conventional methods, TEDUO operates on readily available, unlabeled datasets and addresses the challenge of generalization to previously unseen goals and states. Our approach harnesses large language models (LLMs) in a dual capacity: first, as automatization tools augmenting offline datasets with richer annotations, and second, as generalizable instruction-following agents. Empirical results demonstrate that TEDUO achieves data-efficient learning of robust language-conditioned policies, accomplishing tasks beyond the reach of conventional RL frameworks or out-of-the-box LLMs alone.
Authors: Ruijie Zheng, Yongyuan Liang, Shuaiyi Huang, Jianfeng Gao, Hal Daum\'e III, Andrey Kolobov, Furong Huang, Jianwei Yang
Abstract: Although large vision-language-action (VLA) models pretrained on extensive robot datasets offer promising generalist policies for robotic learning, they still struggle with spatial-temporal dynamics in interactive robotics, making them less effective in handling complex tasks, such as manipulation. In this work, we introduce visual trace prompting, a simple yet effective approach to facilitate VLA models' spatial-temporal awareness for action prediction by encoding state-action trajectories visually. We develop a new TraceVLA model by finetuning OpenVLA on our own collected dataset of 150K robot manipulation trajectories using visual trace prompting. Evaluations of TraceVLA across 137 configurations in SimplerEnv and 4 tasks on a physical WidowX robot demonstrate state-of-the-art performance, outperforming OpenVLA by 10% on SimplerEnv and 3.5x on real-robot tasks and exhibiting robust generalization across diverse embodiments and scenarios. To further validate the effectiveness and generality of our method, we present a compact VLA model based on 4B Phi-3-Vision, pretrained on the Open-X-Embodiment and finetuned on our dataset, rivals the 7B OpenVLA baseline while significantly improving inference efficiency.
Authors: Zhendong Mi, Renming Zheng, Haowen Zhong, Yue Sun, Seth Kneeland, Sayan Moitra, Ken Kutzer, Zhaozhuo Xu Shaoyi Huang
Abstract: Recent advances in agentic LLMs have demonstrated great capabilities in Verilog code generation. However, existing approaches either use LLM-assisted single-agent prompting or cooperation-only multi-agent learning, which will lead to: (i) Degeneration issue for single-agent learning: characterized by diminished error detection and correction capabilities; (ii) Error propagation in cooperation-only multi-agent learning: erroneous information from the former agent will be propagated to the latter through prompts, which can make the latter agents generate buggy code. In this paper, we propose an LLM-based coopetitive multi-agent prompting framework, in which the agents cannot collaborate with each other to form the generation pipeline, but also create a healthy competitive mechanism to improve the generating quality. Our experimental results show that the coopetitive multi-agent framework can effectively mitigate the degeneration risk and reduce the error propagation while improving code error correction capabilities, resulting in higher quality Verilog code generation. The effectiveness of our approach is validated through extensive experiments. On VerilogEval Machine and Human dataset, CoopetitiveV+GPT-4 achieves 99.2% and 99.1% pass@10 scores, respectively. While on RTLLM, CoopetitiveV+GPT-4 obtains 100% syntax and 99.9% functionality pass@5 scores.
Authors: Semih Cayci
Abstract: We analyze the convergence of Gauss-Newton dynamics for training neural networks with smooth activation functions. In the underparameterized regime, the Gauss-Newton gradient flow induces a Riemannian gradient flow on a low-dimensional, smooth, embedded submanifold of the Euclidean output space. Using tools from Riemannian optimization, we prove \emph{last-iterate} convergence of the Riemannian gradient flow to the optimal in-class predictor at an \emph{exponential rate} that is independent of the conditioning of the Gram matrix, \emph{without} requiring explicit regularization. We further characterize the critical impacts of the neural network scaling factor and the initialization on the convergence behavior. In the overparameterized regime, we show that the Levenberg-Marquardt dynamics with an appropriately chosen damping schedule yields fast convergence rate despite potentially ill-conditioned neural tangent kernel matrices, analogous to the underparameterized regime. These findings demonstrate the potential of Gauss-Newton methods for efficiently optimizing neural networks in the near-initialization regime, particularly in ill-conditioned problems where kernel and Gram matrices have small singular values.
Authors: Zhuohao Yu, Weizheng Gu, Yidong Wang, Xingru Jiang, Zhengran Zeng, Jindong Wang, Wei Ye, Shikun Zhang
Abstract: Large Language Models excel at code generation yet struggle with complex programming tasks that demand sophisticated reasoning. To bridge this gap, traditional process supervision relies on learned reward models requiring costly training data and suffering from reward misalignment, while outcome supervision fails for complex tasks needing coordinated intermediate steps. We introduce Outcome Refining Process Supervision, which unifies process and outcome supervision by leveraging executable verification: a tree-structured search framework generates strategic alternatives, profiles execution metrics, and scores candidates via self-critique mechanisms that integrate runtime feedback with reasoning. Experiments across 5 models and 3 benchmarks show consistent gains, with 26.9% higher correctness and 42.2% improved code efficiency. The results demonstrate that ORPS enables LLMs to overcome local optima in code generation, suggesting a promising direction for combining verifiable outcomes with structured reasoning to tackle complex challenges. We open-source at: https://github.com/zhuohaoyu/ORPS
Authors: Wei Liu, Junlong Li, Xiwen Zhang, Fan Zhou, Yu Cheng, Junxian He
Abstract: Self-evolving trainin--where models iteratively learn from their own outputs--has emerged as a key approach for complex reasoning tasks, addressing the scarcity of high-quality chain-of-thought data. However, its effectiveness in multimodal reasoning, a domain more intricate than text-only reasoning, remains underexplored, and the understanding of critical factors in this training paradigm remains limited. Furthermore, a central challenge for this training method is performance saturation, which impedes further improvements and scalability. Inspired by reinforcement learning (RL), in this paper, we reframe self-evolving training for multimodal reasoning through the lens of RL, identifying three pivotal factors: Training Method, Reward Model, and Prompt Variation. Through systematic analysis, we establish relatively optimal design principles that significantly enhance multimodal reasoning capabilities. Moreover, delving deeper into training dynamics, we uncover the roots of saturation and propose a new automatic balancing mechanism to mitigate this limitation. Building on these insights, we propose M-STAR (Multimodal Self-evolving Training for Reasoning), a framework that achieves consistent performance gains across models of varying sizes and diverse benchmarks. All resources are made publicly available at https://mstar-lmm.github.io.
Authors: Simon Krogmann, Pascal Lenzner, Alexander Skopalik
Abstract: We study strategic location choice by customers and sellers, termed the Bakers and Millers Game in the literature. In our generalized setting, each miller can freely choose any location for setting up a mill, while each baker is restricted in the choice of location for setting up a bakery. For optimal bargaining power, a baker would like to select a location with many millers to buy flour from and with little competition from other bakers. Likewise, a miller aims for a location with many bakers and few competing millers. Thus, both types of agents choose locations to optimize the ratio of agents of opposite type divided by agents of the same type at their chosen location. Originally raised in the context of Fractional Hedonic Games, the Bakers and Millers Game has applications that range from commerce to product design. We study the impact of location restrictions on the properties of the game. While pure Nash equilibria trivially exist in the setting without location restrictions, we show via a sophisticated, efficient algorithm that even the more challenging restricted setting admits equilibria. Moreover, the computed equilibrium approximates the optimal social welfare by a factor of at most $2\left(\frac{e}{e-1}\right)$. Furthermore, we give tight bounds on the price of anarchy/stability. On the conceptual side, the location choice feature adds a new layer to the standard setting of Hedonic Games, in the sense that agents that select the same location form a coalition. This allows to naturally restrict the possible coalitions that can be formed. With this, our model generalizes simple symmetric Fractional Hedonic Games on complete bipartite valuation graphs and also Hedonic Diversity Games with utilities single-peaked at 0. We believe that this generalization is also a very interesting direction for other types of Hedonic Games.
Authors: Zhiyuan Fu, Junfan Chen, Lan Zhang, Ting Yang, Jun Niu, Hongyu Sun, Ruidong Li, Peng Liu, Yuqing Zhang
Abstract: Large Language Models (LLMs) are rapidly transforming the landscape of digital content creation. However, the prevalent black-box Application Programming Interface (API) access to many LLMs introduces significant challenges in accountability, governance, and security. LLM fingerprinting, which aims to identify the source model by analyzing statistical and stylistic features of generated text, offers a potential solution. Current progress in this area is hindered by a lack of dedicated datasets and the need for efficient, practical methods that are robust against adversarial manipulations. To address these challenges, we introduce FD-Dataset, a comprehensive bilingual fingerprinting benchmark comprising 90,000 text samples from 20 famous proprietary and open-source LLMs. Furthermore, we present FDLLM, a novel fingerprinting method that leverages parameter-efficient Low-Rank Adaptation (LoRA) to fine-tune a foundation model. This approach enables LoRA to extract deep, persistent features that characterize each source LLM. Through our analysis, we find that LoRA adaptation promotes the aggregation of outputs from the same LLM in representation space while enhancing the separation between different LLMs. This mechanism explains why LoRA proves particularly effective for LLM fingerprinting. Extensive empirical evaluations on FD-Dataset demonstrate FDLLM's superiority, achieving a Macro F1 score 22.1% higher than the strongest baseline. FDLLM also exhibits strong generalization to newly released models, achieving an average accuracy of 95% on unseen models. Notably, FDLLM remains consistently robust under various adversarial attacks, including polishing, translation, and synonym substitution. Experimental results show that FDLLM reduces the average attack success rate from 49.2% (LM-D) to 23.9%.
Authors: Haoxiong Liu, Jiacheng Sun, Zhenguo Li, Andrew C Yao
Abstract: The synergy between deep learning models and traditional automation tools, such as built-in tactics of the proof assistant and off-the-shelf automated theorem provers, plays a crucial role in developing robust and efficient neural theorem provers(NTPs). However, for proof synthesis with LLMs, previous work applies automation tools either only when explicitly invoked by the model or at a single granularity level, failing to fully exploit their power. To solve this issue, we propose ProofAug, a procedure that equips LLMs with automation methods at various granularities through fine-grained structure analysis of model-generated proof proposals. ProofAug also serves as a versatile plug-and-play module that seamlessly integrates with any tree-search algorithm, enabling our construction of an efficient recursive proving (ERP) module to further enhance performance. The superiority of our method is validated on the miniF2F benchmark using the open-source deepseek-math-7b-base model and the Isabelle proof assistant. Notably, by additionally employing a mixed prompting strategy, we achieve a cumulative pass rate of 66.0% after curation of the dataset (61.9% for the original version) with 2100 queries to the model per problem (In contrast, the previous SOTA in Isabelle, Subgoal-XL, only achieves 56.1% using 16384 queries per problem). We also implement a Lean 4 version of ProofAug that can improve the pass@1 performance of Kimina-Prover-Preview-Distill-1.5B from 44.3% to 50.4% on miniF2F-test. Our code is available at https://github.com/haoxiongliu/ProofAug.
Authors: Mohammad Fatahi, Danial Sadrian Zadeh, Benyamin Ghojogh, Behzad Moshiri, Otman Basir
Abstract: Intelligent transportation systems (ITS) play a pivotal role in modern infrastructure but face security risks due to the broadcast-based nature of the in-vehicle Controller Area Network (CAN) buses. While numerous machine learning models and strategies have been proposed to detect CAN anomalies, existing approaches lack robustness evaluations and fail to comprehensively detect attacks due to shifting their focus on a subset of dominant structures of anomalies. To overcome these limitations, the current study proposes a cascade feature-level spatiotemporal fusion framework that integrates the spatial features and temporal features through a two-parameter genetic algorithm (2P-GA)-optimized cascade architecture to cover all dominant structures of anomalies. Extensive paired t-test analysis confirms that the model achieves an AUC-ROC of 0.9987, demonstrating robust anomaly detection capabilities. The Spatial Module improves the precision by approximately 4%, while the Temporal Module compensates for recall losses, ensuring high true positive rates. The proposed framework detects all attack types with 100% accuracy on the CAR-HACKING dataset, outperforming state-of-the-art methods. This study provides a validated, robust solution for real-world CAN security challenges.
Authors: Isha Puri, Shivchander Sudalairaj, Guangxuan Xu, Kai Xu, Akash Srivastava
Abstract: Large language models (LLMs) have achieved significant performance gains via scaling up model sizes and/or data. However, recent evidence suggests diminishing returns from such approaches, motivating scaling the computation spent at inference time. Existing inference-time scaling methods, usually with reward models, cast the task as a search problem, which tends to be vulnerable to reward hacking as a consequence of approximation errors in reward models. In this paper, we instead cast inference-time scaling as a probabilistic inference task and leverage sampling-based techniques to explore the typical set of the state distribution of a state-space model with an approximate likelihood, rather than optimize for its mode directly. We propose a novel inference-time scaling approach by adapting particle-based Monte Carlo methods to this task. Our empirical evaluation demonstrates that our methods have a 4-16x better scaling rate over our deterministic search counterparts on various challenging mathematical reasoning tasks. Using our approach, we show that Qwen2.5-Math-1.5B-Instruct can surpass GPT-4o accuracy in only 4 rollouts, while Qwen2.5-Math-7B-Instruct scales to o1 level accuracy in only 32 rollouts. Our work not only presents an effective method to inference-time scaling, but also connects the rich literature in probabilistic inference with inference-time scaling of LLMs to develop more robust algorithms in future work. Code, videos, and further information available at https://probabilistic-inference-scaling.github.io.
Authors: Jeonghoon Kim, Byeongchan Lee, Cheonbok Park, Yeontaek Oh, Beomjun Kim, Taehwan Yoo, Seongjin Shin, Dongyoon Han, Jinwoo Shin, Kang Min Yoo
Abstract: Selecting a layer normalization (LN) strategy that stabilizes training and speeds convergence in Transformers remains difficult, even for today's large language models (LLM). We present a comprehensive analytical foundation for understanding how different LN strategies influence training dynamics in large-scale Transformers. Until recently, Pre-LN and Post-LN have long dominated practices despite their limitations in large-scale training. However, several open-source models have recently begun silently adopting a third strategy without much explanation. This strategy places normalization layer peripherally around sublayers, a design we term Peri-LN. While Peri-LN has demonstrated promising performance, its precise mechanisms and benefits remain almost unexplored. Our in-depth analysis delineates the distinct behaviors of LN strategies, showing how each placement shapes activation variance and gradient propagation. To validate our theoretical insight, we conduct extensive experiments on Transformers up to $3.2$B parameters, showing that Peri-LN consistently achieves more balanced variance growth, steadier gradient flow, and convergence stability. Our results suggest that Peri-LN warrants broader consideration for large-scale Transformer architectures, providing renewed insights into the optimal placement of LN.
Authors: Akash Kumar
Abstract: The success of deep networks is crucially attributed to their ability to capture latent features within a representation space. In this work, we investigate whether the underlying learned features of a model can be efficiently retrieved through feedback from an agent, such as a large language model (LLM), in the form of relative \textit{triplet comparisons}. These features may represent various constructs, including dictionaries in LLMs or a covariance matrix of Mahalanobis distances. We analyze the feedback complexity associated with learning a feature matrix in sparse settings. Our results establish tight bounds when the agent is permitted to construct activations and demonstrate strong upper bounds in sparse scenarios when the agent's feedback is limited to distributional information. We validate our theoretical findings through experiments on two distinct applications: feature recovery from Recursive Feature Machines and dictionary extraction from sparse autoencoders trained on Large Language Models.
Authors: Kaizhen Zhu, Mokai Pan, Yuexin Ma, Yanwei Fu, Jingyi Yu, Jingya Wang, Ye Shi
Abstract: Recent advances in diffusion bridge models leverage Doob's $h$-transform to establish fixed endpoints between distributions, demonstrating promising results in image translation and restoration tasks. However, these approaches frequently produce blurred or excessively smoothed image details and lack a comprehensive theoretical foundation to explain these shortcomings. To address these limitations, we propose UniDB, a unified framework for diffusion bridges based on Stochastic Optimal Control (SOC). UniDB formulates the problem through an SOC-based optimization and derives a closed-form solution for the optimal controller, thereby unifying and generalizing existing diffusion bridge models. We demonstrate that existing diffusion bridges employing Doob's $h$-transform constitute a special case of our framework, emerging when the terminal penalty coefficient in the SOC cost function tends to infinity. By incorporating a tunable terminal penalty coefficient, UniDB achieves an optimal balance between control costs and terminal penalties, substantially improving detail preservation and output quality. Notably, UniDB seamlessly integrates with existing diffusion bridge models, requiring only minimal code modifications. Extensive experiments across diverse image restoration tasks validate the superiority and adaptability of the proposed framework. Our code is available at https://github.com/UniDB-SOC/UniDB/.
Authors: Quan Wei, Chung-Yiu Yau, Hoi-To Wai, Yang Katie Zhao, Dongyeop Kang, Youngsuk Park, Mingyi Hong
Abstract: Supervised fine-tuning is a standard method for adapting pre-trained large language models (LLMs) to downstream tasks. Quantization has been recently studied as a post-training technique for efficient LLM deployment. To obtain quantized fine-tuned LLMs, conventional pipelines would first fine-tune the pre-trained models, followed by post-training quantization. This often yields suboptimal performance as it fails to leverage the synergy between fine-tuning and quantization. To effectively realize low-bit quantization of weights, activations and KV caches in LLMs, we propose an algorithm named Rotated Straight-Through-Estimator (RoSTE), which combines quantization-aware supervised fine-tuning (QA-SFT) with an adaptive rotation strategy that identifies an effective rotation configuration to reduce activation outliers. We provide theoretical insights on RoSTE by analyzing its prediction error when applied to an overparameterized least square quantized training problem. Our findings reveal that the prediction error is directly proportional to the quantization error of the converged weights, which can be effectively managed through an optimized rotation configuration. Experiments on Pythia, Qwen and Llama models of different sizes demonstrate the effectiveness of RoSTE. Compared to existing post-SFT quantization baselines, our method consistently achieves superior performances across various tasks and different LLM architectures. Our code is available at https://github.com/OptimAI-Lab/RoSTE.
Authors: Andrea Cini, Alexander Jenkins, Danilo Mandic, Cesare Alippi, Filippo Maria Bianchi
Abstract: We address the problem of uncertainty quantification in time series forecasting by exploiting observations at correlated sequences. Relational deep learning methods leveraging graph representations are among the most effective tools for obtaining point estimates from spatiotemporal data and correlated time series. However, the problem of exploiting relational structures to estimate the uncertainty of such predictions has been largely overlooked in the same context. To this end, we propose a novel distribution-free approach based on the conformal prediction framework and quantile regression. Despite the recent applications of conformal prediction to sequential data, existing methods operate independently on each target time series and do not account for relationships among them when constructing the prediction interval. We fill this void by introducing a novel conformal prediction method based on graph deep learning operators. Our approach, named Conformal Relational Prediction (CoRel), does not require the relational structure (graph) to be known a priori and can be applied on top of any pre-trained predictor. Additionally, CoRel includes an adaptive component to handle non-exchangeable data and changes in the input time series. Our approach provides accurate coverage and achieves state-of-the-art uncertainty quantification in relevant benchmarks.
Authors: Shahriar Kabir Nahin, Rabindra Nath Nandi, Sagor Sarker, Quazi Sarwar Muhtaseem, Md Kowsher, Apu Chandraw Shill, Md Ibrahim, Mehadi Hasan Menon, Tareq Al Muntasir, Firoj Alam
Abstract: In this paper, we present TituLLMs, the first large pretrained Bangla LLMs, available in 1b and 3b parameter sizes. Due to computational constraints during both training and inference, we focused on smaller models. To train TituLLMs, we collected a pretraining dataset of approximately ~37 billion tokens. We extended the Llama-3.2 tokenizer to incorporate language- and culture-specific knowledge, which also enables faster training and inference. There was a lack of benchmarking datasets to benchmark LLMs for Bangla. To address this gap, we developed five benchmarking datasets. We benchmarked various LLMs, including TituLLMs, and demonstrated that TituLLMs outperforms its initial multilingual versions. However, this is not always the case, highlighting the complexities of language adaptation. Our work lays the groundwork for adapting existing multilingual open models to other low-resource languages. To facilitate broader adoption and further research, we have made the TituLLMs models and benchmarking datasets publicly available (https://huggingface.co/collections/hishab/titulm-llama-family-6718d31fc1b83529276f490a).
URLs: https://huggingface.co/collections/hishab/titulm-llama-family-6718d31fc1b83529276f490a).
Authors: Xiaoyi Dong, Jian Cheng, Xi Sheryl Zhang
Abstract: The Soft Actor-Critic (SAC) algorithm with a Gaussian policy has become a mainstream implementation for realizing the Maximum Entropy Reinforcement Learning (MaxEnt RL) objective, which incorporates entropy maximization to encourage exploration and enhance policy robustness. While the Gaussian policy performs well on simpler tasks, its exploration capacity and potential performance in complex multi-goal RL environments are limited by its inherent unimodality. In this paper, we employ the diffusion model, a powerful generative model capable of capturing complex multimodal distributions, as the policy representation to fulfill the MaxEnt RL objective, developing a method named MaxEnt RL with Diffusion Policy (MaxEntDP). Our method enables efficient exploration and brings the policy closer to the optimal MaxEnt policy. Experimental results on Mujoco benchmarks show that MaxEntDP outperforms the Gaussian policy and other generative models within the MaxEnt RL framework, and performs comparably to other state-of-the-art diffusion-based online RL algorithms. Our code is available at https://github.com/diffusionyes/MaxEntDP.
Authors: Prasanna Mayilvahanan, Thadd\"aus Wiedemer, Sayak Mallick, Matthias Bethge, Wieland Brendel
Abstract: Scaling laws guide the development of large language models (LLMs) by offering estimates for the optimal balance of model size, tokens, and compute. More recently, loss-to-loss scaling laws that relate losses across pretraining datasets and downstream tasks have emerged as a powerful tool for understanding and improving LLM performance. In this work, we investigate which factors most strongly influence loss-to-loss scaling. Our experiments reveal that the pretraining data and tokenizer determine the scaling trend. In contrast, model size, optimization hyperparameters, and even significant architectural differences, such as between transformer-based models like Llama and state-space models like Mamba, have limited impact. Consequently, practitioners should carefully curate suitable pretraining datasets for optimal downstream performance, while architectures and other settings can be freely optimized for training efficiency.
Authors: Changhoon Kim, Yanjun Qi
Abstract: Text-to-Image (T2I) models have made remarkable progress in generating high-quality, diverse visual content from natural language prompts. However, their ability to reproduce copyrighted styles, sensitive imagery, and harmful content raises significant ethical and legal concerns. Concept erasure offers a proactive alternative to external filtering by modifying T2I models to prevent the generation of undesired content. In this survey, we provide a structured overview of concept erasure, categorizing existing methods based on their optimization strategies and the architectural components they modify. We categorize concept erasure methods into fine-tuning for parameter updates, closed-form solutions for efficient edits, and inference-time interventions for content restriction without weight modification. Additionally, we explore adversarial attacks that bypass erasure techniques and discuss emerging defenses. To support further research, we consolidate key datasets, evaluation metrics, and benchmarks for assessing erasure effectiveness and model robustness. This survey serves as a comprehensive resource, offering insights into the evolving landscape of concept erasure, its challenges, and future directions.
Authors: Hengzhi Li, Megan Tjandrasuwita, Yi R. Fung, Armando Solar-Lezama, Paul Pu Liang
Abstract: As AI becomes more closely integrated with peoples' daily activities, socially intelligent AI that can understand and interact seamlessly with humans in daily lives is increasingly important. However, current works in AI social reasoning all rely on language-only or language-dominant approaches to benchmark and training models, resulting in systems that are improving in verbal communication but struggle with nonverbal social understanding. To address this limitation, we tap into a novel data source rich in nonverbal social interactions -- mime videos. Mimes refer to the art of expression through gesture and movement without spoken words, which presents unique challenges and opportunities in interpreting nonverbal social communication. We contribute a new dataset called MimeQA, obtained by sourcing 8 hours of videos clips from YouTube and developing a comprehensive video question-answering benchmark comprising 806 carefully annotated and verified question-answer pairs, designed to probe nonverbal social reasoning capabilities. Using MimeQA, we evaluate state-of-the-art video large language models (vLLMs) and find that they achieve low overall accuracy, ranging from 20-30%, while humans score 86%. Our analysis reveals that vLLMs often fail to ground imagined objects and over-rely on the text prompt while ignoring subtle nonverbal interactions. We hope to inspire future work in AI models that embody true social intelligence capable of interpreting non-verbal human interactions.
Authors: Jintao Zhang, Chendong Xiang, Haofeng Huang, Jia Wei, Haocheng Xi, Jun Zhu, Jianfei Chen
Abstract: An efficient attention implementation is essential for large models due to its quadratic time complexity. Fortunately, attention commonly exhibits sparsity, i.e., many values in the attention map are near zero, allowing for the omission of corresponding computations. Many studies have utilized the sparse pattern to accelerate attention. However, most existing works focus on optimizing attention within specific models by exploiting certain sparse patterns of the attention map. A universal sparse attention that guarantees both the speedup and end-to-end performance of diverse models remains elusive. In this paper, we propose SpargeAttn, a universal sparse and quantized attention for any model. Our method uses a two-stage online filter: in the first stage, we rapidly and accurately predict the attention map, enabling the skip of some matrix multiplications in attention. In the second stage, we design an online softmax-aware filter that incurs no extra overhead and further skips some matrix multiplications. Experiments show that our method significantly accelerates diverse models, including language, image, and video generation, without sacrificing end-to-end metrics. The codes are available at https://github.com/thu-ml/SpargeAttn.
Authors: Lucy Farnik, Tim Lawson, Conor Houghton, Laurence Aitchison
Abstract: Sparse autoencoders (SAEs) have been successfully used to discover sparse and human-interpretable representations of the latent activations of LLMs. However, we would ultimately like to understand the computations performed by LLMs and not just their representations. The extent to which SAEs can help us understand computations is unclear because they are not designed to "sparsify" computations in any sense, only latent activations. To solve this, we propose Jacobian SAEs (JSAEs), which yield not only sparsity in the input and output activations of a given model component but also sparsity in the computation (formally, the Jacobian) connecting them. With a na\"ive implementation, the Jacobians in LLMs would be computationally intractable due to their size. One key technical contribution is thus finding an efficient way of computing Jacobians in this setup. We find that JSAEs extract a relatively large degree of computational sparsity while preserving downstream LLM performance approximately as well as traditional SAEs. We also show that Jacobians are a reasonable proxy for computational sparsity because MLPs are approximately linear when rewritten in the JSAE basis. Lastly, we show that JSAEs achieve a greater degree of computational sparsity on pre-trained LLMs than on the equivalent randomized LLM. This shows that the sparsity of the computational graph appears to be a property that LLMs learn through training, and suggests that JSAEs might be more suitable for understanding learned transformer computations than standard SAEs.
Authors: Lei Zhao, Sizhou Chen, Linfeng Feng, Jichao Zhang, Xiao-Lei Zhang, Chi Zhang, Xuelong Li
Abstract: Text-to-audio (TTA), which generates audio signals from textual descriptions, has received huge attention in recent years. However, recent works focused on text to monaural audio only. As we know, spatial audio provides more immersive auditory experience than monaural audio, e.g. in virtual reality. To address this issue, we propose a text-to-spatial-audio (TTSA) generation framework named DualSpec. Specifically, it first trains variational autoencoders (VAEs) for extracting the latent acoustic representations from sound event audio. Then, given text that describes sound events and event directions, the proposed method uses the encoder of a pretrained large language model to transform the text into text features. Finally, it trains a diffusion model from the latent acoustic representations and text features for the spatial audio generation. In the inference stage, only the text description is needed to generate spatial audio. Particularly, to improve the synthesis quality and azimuth accuracy of the spatial sound events simultaneously, we propose to use two kinds of acoustic features. One is the Mel spectrograms which is good for improving the synthesis quality, and the other is the short-time Fourier transform spectrograms which is good at improving the azimuth accuracy. We provide a pipeline of constructing spatial audio dataset with text prompts, for the training of the VAEs and diffusion model. We also introduce new spatial-aware evaluation metrics to quantify the azimuth errors of the generated spatial audio recordings. Experimental results demonstrate that the proposed method can generate spatial audio with high directional and event consistency.
Authors: Bojana Ba\v{s}aragin, Darija Medvecki, Gorana Goji\'c, Milena Oparnica, Dragi\v{s}a Mi\v{s}kovi\'c
Abstract: This study introduces a novel natural language processing pipeline that enhances customer service efficiency at Telekom Srbija, a leading Serbian telecommunications company, through automated email topic detection and labeling. Central to the pipeline is BERTopic, a modular framework that allows unsupervised topic modeling. After a series of preprocessing and postprocessing steps, we assign one of 12 topics and several additional labels to incoming emails, allowing customer service to filter and access them through a custom-made application. While applied to Serbian, the methodology is conceptually language-agnostic and can be readily adapted to other languages, particularly those that are low-resourced and morphologically rich. The system performance was evaluated by assessing the speed and correctness of the automatically assigned topics, with a weighted average processing time of 0.041 seconds per email and a weighted average F1 score of 0.96. The system now operates in the company's production environment, streamlining customer service operations through automated email classification.
Authors: Yukang Yang, Declan Campbell, Kaixuan Huang, Mengdi Wang, Jonathan Cohen, Taylor Webb
Abstract: Many recent studies have found evidence for emergent reasoning capabilities in large language models (LLMs), but debate persists concerning the robustness of these capabilities, and the extent to which they depend on structured reasoning mechanisms. To shed light on these issues, we study the internal mechanisms that support abstract reasoning in LLMs. We identify an emergent symbolic architecture that implements abstract reasoning via a series of three computations. In early layers, symbol abstraction heads convert input tokens to abstract variables based on the relations between those tokens. In intermediate layers, symbolic induction heads perform sequence induction over these abstract variables. Finally, in later layers, retrieval heads predict the next token by retrieving the value associated with the predicted abstract variable. These results point toward a resolution of the longstanding debate between symbolic and neural network approaches, suggesting that emergent reasoning in neural networks depends on the emergence of symbolic mechanisms.
Authors: Jiawei Zhang, Xuan Yang, Taiqi Wang, Yu Yao, Aleksandr Petiushko, Bo Li
Abstract: Traditional autonomous driving systems often struggle to connect high-level reasoning with low-level control, leading to suboptimal and sometimes unsafe behaviors. Recent advances in multimodal large language models (MLLMs), which process both visual and textual data, offer an opportunity to unify perception and reasoning. However, effectively embedding precise safety knowledge into MLLMs for autonomous driving remains a significant challenge. To address this, we propose SafeAuto, a framework that enhances MLLM-based autonomous driving by incorporating both unstructured and structured knowledge. First, we introduce a Position-Dependent Cross-Entropy (PDCE) loss to improve low-level control signal predictions when values are represented as text. Second, to explicitly integrate safety knowledge, we develop a reasoning component that translates traffic rules into first-order logic (e.g., "red light $\implies$ stop") and embeds them into a probabilistic graphical model (e.g., Markov Logic Network) to verify predicted actions using recognized environmental attributes. Additionally, our Multimodal Retrieval-Augmented Generation (RAG) model leverages video, control signals, and environmental attributes to learn from past driving experiences. Integrating PDCE, MLN, and Multimodal RAG, SafeAuto outperforms existing baselines across multiple datasets, enabling more accurate, reliable, and safer autonomous driving. The code is available at https://github.com/AI-secure/SafeAuto.
Authors: Wei Liu, Tao Zhang, Chenhui Lin, Kaiwen Li, Rui Wang
Abstract: Independent microgrids are crucial for supplying electricity by combining distributed energy resources and loads in scenarios like isolated islands and field combat. Fast and accurate assessments of microgrid vulnerability against intentional attacks or natural disasters are essential for effective risk prevention and design optimization. However, conventional Monte Carlo simulation (MCS) methods are computationally expensive and time-consuming, while existing machine learning-based approaches often lack accuracy and explainability. To address these challenges, this study proposes a fast and explainable vulnerability assessment framework that integrates MCS with a graph attention network enhanced by self-attention pooling (GAT-S). MCS generates training data, while the GAT-S model learns the structural and electrical characteristics of the microgrid and further assesses its vulnerability intelligently. The GAT-S improves explainability and computational efficiency by dynamically assigning attention weights to critical nodes. Comprehensive experimental evaluations across various microgrid configurations demonstrate that the proposed framework provides accurate vulnerability assessments, achieving a mean squared error as low as 0.001, real-time responsiveness within 1 second, and delivering explainable results.
Authors: Jintao Zhang, Guoliang Li, Jinyang Su
Abstract: Retrieval-augmented generation (RAG) has demonstrated significant proficiency in conducting question-answering (QA) tasks within a specified corpus. Nonetheless, numerous failure instances of RAG in QA still exist. These failures are not solely attributable to the limitations of Large Language Models (LLMs); instead, they predominantly arise from the retrieval of inaccurate information for LLMs due to two limitations: (1) Current RAG methods segment the corpus without considering semantics, making it difficult to find relevant context due to impaired correlation between questions and the segments. (2) There is a trade-off between missing essential context with fewer context retrieved and getting irrelevant context with more context retrieved. In this paper, we introduce a RAG framework (SAGE), to overcome these limitations. First, to address the segmentation issue without considering semantics, we propose to train a semantic segmentation model. This model is trained to segment the corpus into semantically complete chunks. Second, to ensure that only the most relevant chunks are retrieved while the irrelevant ones are ignored, we design a chunk selection algorithm to dynamically select chunks based on the decreasing speed of the relevance score, leading to a more relevant selection. Third, to further ensure the precision of the retrieved chunks, we propose letting LLMs assess whether retrieved chunks are excessive or lacking and then adjust the amount of context accordingly. Experiments show that SAGE outperforms baselines by 61.25% in the quality of QA on average. Moreover, by avoiding retrieving noisy context, SAGE lowers the cost of the tokens consumed in LLM inference and achieves a 49.41% enhancement in cost efficiency on average. Additionally, our work offers valuable insights for boosting RAG.
Authors: Jiawei Zhang, Shuang Yang, Bo Li
Abstract: Large Language Model (LLM) agents equipped with external tools have become increasingly powerful for complex tasks such as web shopping, automated email replies, and financial trading. However, these advancements amplify the risks of adversarial attacks, especially when agents can access sensitive external functionalities. Nevertheless, manipulating LLM agents into performing targeted malicious actions or invoking specific tools remains challenging, as these agents extensively reason or plan before executing final actions. In this work, we present UDora, a unified red teaming framework designed for LLM agents that dynamically hijacks the agent's reasoning processes to compel malicious behavior. Specifically, UDora first generates the model's reasoning trace for the given task, then automatically identifies optimal points within this trace to insert targeted perturbations. The resulting perturbed reasoning is then used as a surrogate response for optimization. By iteratively applying this process, the LLM agent will then be induced to undertake designated malicious actions or to invoke specific malicious tools. Our approach demonstrates superior effectiveness compared to existing methods across three LLM agent datasets. The code is available at https://github.com/AI-secure/UDora.
Authors: Renato Lui Geh, Zilei Shao, Guy Van den Broeck
Abstract: Current LLM pipelines account for only one possible tokenization for a given string, ignoring exponentially many alternative tokenizations during training and inference. For example, the standard Llama3 tokenization of penguin is [p,enguin], yet [peng,uin] is another perfectly valid alternative. In this paper, we show that despite LLMs being trained solely on one tokenization, they still retain semantic understanding of other tokenizations, raising questions about their implications in LLM safety. Put succinctly, we answer the following question: can we adversarially tokenize an obviously malicious string to evade safety and alignment restrictions? We show that not only is adversarial tokenization an effective yet previously neglected axis of attack, but it is also competitive against existing state-of-the-art adversarial approaches without changing the text of the harmful request. We empirically validate this exploit across three state-of-the-art LLMs and adversarial datasets, revealing a previously unknown vulnerability in subword models.
Authors: Yixiang Sun, Haotian Fu, Michael Littman, George Konidaris
Abstract: We propose DRAGO, a novel approach for continual model-based reinforcement learning aimed at improving the incremental development of world models across a sequence of tasks that differ in their reward functions but not the state space or dynamics. DRAGO comprises two key components: Synthetic Experience Rehearsal, which leverages generative models to create synthetic experiences from past tasks, allowing the agent to reinforce previously learned dynamics without storing data, and Regaining Memories Through Exploration, which introduces an intrinsic reward mechanism to guide the agent toward revisiting relevant states from prior tasks. Together, these components enable the agent to maintain a comprehensive and continually developing world model, facilitating more effective learning and adaptation across diverse environments. Empirical evaluations demonstrate that DRAGO is able to preserve knowledge across tasks, achieving superior performance in various continual learning scenarios.
Authors: Dong Shu, Xuansheng Wu, Haiyan Zhao, Daking Rai, Ziyu Yao, Ninghao Liu, Mengnan Du
Abstract: Large Language Models (LLMs) have transformed natural language processing, yet their internal mechanisms remain largely opaque. Recently, mechanistic interpretability has attracted significant attention from the research community as a means to understand the inner workings of LLMs. Among various mechanistic interpretability approaches, Sparse Autoencoders (SAEs) have emerged as a promising method due to their ability to disentangle the complex, superimposed features within LLMs into more interpretable components. This paper presents a comprehensive survey of SAEs for interpreting and understanding the internal workings of LLMs. Our major contributions include: (1) exploring the technical framework of SAEs, covering basic architecture, design improvements, and effective training strategies; (2) examining different approaches to explaining SAE features, categorized into input-based and output-based explanation methods; (3) discussing evaluation methods for assessing SAE performance, covering both structural and functional metrics; and (4) investigating real-world applications of SAEs in understanding and manipulating LLM behaviors.
Authors: Jianwen Sun, Yukang Feng, Chuanhao Li, Fanrui Zhang, Zizhen Li, Jiaxin Ai, Sizhuo Zhou, Yu Dai, Shenglin Zhang, Kaipeng Zhang
Abstract: Unified multimodal understanding and generation have recently received much attention in the area of vision and language. Existing UniMs are designed to simultaneously learn both multimodal understanding and generation capabilities, demanding substantial computational resources, and often struggle to generate interleaved text-image. We present ARMOR, a resource-efficient and pure autoregressive framework that achieves both understanding and generation by fine-tuning existing multimodal large language models (MLLMs). Specifically, ARMOR extends existing MLLMs from three perspectives: (1) For model architecture, an asymmetric encoder-decoder architecture with a forward-switching mechanism is introduced to unify embedding space integrating textual and visual modalities for enabling natural text-image interleaved generation with minimal computational overhead. (2) For training data, a meticulously curated, high-quality interleaved dataset is collected for fine-tuning MLLMs. (3) For the training algorithm, we propose a ``what or how to generate'' algorithm to empower existing MLLMs with multimodal generation capabilities while preserving their multimodal understanding capabilities, through three progressive training stages based on the collected dataset. Experimental results demonstrate that ARMOR upgrades existing MLLMs to UniMs with promising image generation capabilities, using limited training resources. Our code will be released soon at https://github.com/finyorko/armor.
Authors: Abir Harrasse, Philip Quirke, Clement Neo, Dhruv Nathawani, Luke Marks, Amir Abdullah
Abstract: Mechanistic interpretability research faces a gap between analyzing simple circuits in toy tasks and discovering features in large models. To bridge this gap, we propose text-to-SQL generation as an ideal task to study, as it combines the formal structure of toy tasks with real-world complexity. We introduce TinySQL, a synthetic dataset, progressing from basic to advanced SQL operations, and train models ranging from 33M to 1B parameters to establish a comprehensive testbed for interpretability. We apply multiple complementary interpretability techniques, including Edge Attribution Patching and Sparse Autoencoders, to identify minimal circuits and components supporting SQL generation. We compare circuits for different SQL subskills, evaluating their minimality, reliability, and identifiability. Finally, we conduct a layerwise logit lens analysis to reveal how models compose SQL queries across layers: from intent recognition to schema resolution to structured generation. Our work provides a robust framework for probing and comparing interpretability methods in a structured, progressively complex setting.
Authors: Sungyeon Kim, Xinliang Zhu, Xiaofan Lin, Muhammet Bastan, Douglas Gray, Suha Kwak
Abstract: Generative retrieval is an emerging approach in information retrieval that generates identifiers (IDs) of target data based on a query, providing an efficient alternative to traditional embedding-based retrieval methods. However, existing models are task-specific and fall short of embedding-based retrieval in performance. This paper proposes GENIUS, a universal generative retrieval framework supporting diverse tasks across multiple modalities and domains. At its core, GENIUS introduces modality-decoupled semantic quantization, transforming multimodal data into discrete IDs encoding both modality and semantics. Moreover, to enhance generalization, we propose a query augmentation that interpolates between a query and its target, allowing GENIUS to adapt to varied query forms. Evaluated on the M-BEIR benchmark, it surpasses prior generative methods by a clear margin. Unlike embedding-based retrieval, GENIUS consistently maintains high retrieval speed across database size, with competitive performance across multiple benchmarks. With additional re-ranking, GENIUS often achieves results close to those of embedding-based methods while preserving efficiency.
Authors: Jiztom Kavalakkatt Francis, Matthew J Darr
Abstract: This paper introduces a novel framework that accelerates the discovery of actionable relationships in high-dimensional temporal data by integrating machine learning (ML), explainable AI (XAI), and natural language processing (NLP) to enhance data quality and streamline workflows. Traditional methods often fail to recognize complex temporal relationships, leading to noisy, redundant, or biased datasets. Our approach combines ML-driven pruning to identify and mitigate low-quality samples, XAI-based interpretability to validate critical feature interactions, and NLP for future contextual validation, reducing the time required to uncover actionable insights by 40-60%. Evaluated on real-world agricultural and synthetic datasets, the framework significantly improves performance metrics (e.g., MSE, R2, MAE) and computational efficiency, with hardware-agnostic scalability across diverse platforms. While long-term real-world impacts (e.g., cost savings, sustainability gains) are pending, this methodology provides an immediate pathway to accelerate data-centric AI in dynamic domains like agriculture and energy, enabling faster iteration cycles for domain experts.
Authors: Mateusz Pach, Shyamgopal Karthik, Quentin Bouniot, Serge Belongie, Zeynep Akata
Abstract: Given that interpretability and steerability are crucial to AI safety, Sparse Autoencoders (SAEs) have emerged as a tool to enhance them in Large Language Models (LLMs). In this work, we extend the application of SAEs to Vision-Language Models (VLMs), such as CLIP, and introduce a comprehensive framework for evaluating monosemanticity at the neuron-level in vision representations. To ensure that our evaluation aligns with human perception, we propose a benchmark derived from a large-scale user study. Our experimental results reveal that SAEs trained on VLMs significantly enhance the monosemanticity of individual neurons, with sparsity and wide latents being the most influential factors. Notably, we demonstrate that applying SAE interventions on CLIP's vision encoder directly steers multimodal LLM outputs (e.g., LLaVA), without any modifications to the underlying model. These findings emphasize the practicality and efficacy of SAEs as an unsupervised tool for enhancing both interpretability and control of VLMs. Code is available at https://github.com/ExplainableML/sae-for-vlm.
Authors: Sanchit Kabra, Akshita Jha, Chandan K. Reddy
Abstract: Recent advances in large-scale generative language models have shown that reasoning capabilities can significantly improve model performance across a variety of tasks. However, the impact of reasoning on a model's ability to mitigate stereotypical responses remains largely underexplored. In this work, we investigate the crucial relationship between a model's reasoning ability and fairness, and ask whether improved reasoning capabilities can mitigate harmful stereotypical responses, especially those arising due to shallow or flawed reasoning. We conduct a comprehensive evaluation of multiple open-source LLMs, and find that larger models with stronger reasoning abilities exhibit substantially lower stereotypical bias on existing fairness benchmarks. Building on this insight, we introduce ReGiFT -- Reasoning Guided Fine-Tuning, a novel approach that extracts structured reasoning traces from advanced reasoning models and infuses them into models that lack such capabilities. We use only general-purpose reasoning and do not require any fairness-specific supervision for bias mitigation. Notably, we see that models fine-tuned using ReGiFT not only improve fairness relative to their non-reasoning counterparts but also outperform advanced reasoning models on fairness benchmarks. We also analyze how variations in the correctness of the reasoning traces and their length influence model fairness and their overall performance. Our findings highlight that enhancing reasoning capabilities is an effective, fairness-agnostic strategy for mitigating stereotypical bias caused by reasoning flaws.
Authors: Beilong Tang, Bang Zeng, Ming Li
Abstract: We propose LauraTSE, an Auto-Regressive Decoder-Only Language Model for Target Speaker Extraction built upon the LauraGPT backbone. LauraTSE employs a small-scale auto-regressive decoder-only language model that generates the initial layers of the target speech's discrete codec representations from the continuous embeddings of both the mixture and reference speech. These outputs serve as coarse-grained predictions. To refine them, a one-step encoder-only language model reconstructs the full codec representation by integrating information from both the mixture and the reference speech, adding fine-grained details. Our approach achieves superior or comparable performance to existing TSE models. Additionally, we conduct ablation studies to investigate the data scalability and the contribution of the encoder-only model.
Authors: Yixuan Even Xu, Yash Savani, Fei Fang, Zico Kolter
Abstract: Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for enhancing reasoning capabilities in large language models. However, it is constrained by a fundamental asymmetry in computation and memory requirements: rollout generation is embarrassingly parallel and memory-light, whereas policy updates are communication-heavy and memory-intensive. To address this, we introduce PODS (Policy Optimization with Down-Sampling). PODS produces numerous rollouts in parallel, then trains on only an informative subset, preserving learning signals while slashing update cost. We instantiate PODS with max-variance down-sampling, a principled criterion that maximises reward diversity and show it admits an $O(n\log n)$ solution. Empirically, coupling PODS with Group Relative Policy Optimization (GRPO) achieves superior performance over standard GRPO across different reasoning benchmarks and hardware environments.
Authors: Xinyu Wang, Jijun Chi, Zhenghan Tai, Tung Sum Thomas Kwok, Muzhi Li, Zhuhong Li, Hailin He, Yuchen Hua, Peng Lu, Suyuchen Wang, Yihong Wu, Jerry Huang, Jingrui Tian, Fengran Mo, Yufei Cui, Ling Zhou
Abstract: Leveraging large language models in real-world settings often entails a need to utilize domain-specific data and tools in order to follow the complex regulations that need to be followed for acceptable use. Within financial sectors, modern enterprises increasingly rely on Retrieval-Augmented Generation (RAG) systems to address complex compliance requirements in financial document workflows. However, existing solutions struggle to account for the inherent heterogeneity of data (e.g., text, tables, diagrams) and evolving nature of regulatory standards used in financial filings, leading to compromised accuracy in critical information extraction. We propose the FinSage framework as a solution, utilizing a multi-aspect RAG framework tailored for regulatory compliance analysis in multi-modal financial documents. FinSage introduces three innovative components: (1) a multi-modal pre-processing pipeline that unifies diverse data formats and generates chunk-level metadata summaries, (2) a multi-path sparse-dense retrieval system augmented with query expansion (HyDE) and metadata-aware semantic search, and (3) a domain-specialized re-ranking module fine-tuned via Direct Preference Optimization (DPO) to prioritize compliance-critical content. Extensive experiments demonstrate that FinSage achieves an impressive recall of 92.51% on 75 expert-curated questions derived from surpasses the best baseline method on the FinanceBench question answering datasets by 24.06% in accuracy. Moreover, FinSage has been successfully deployed as financial question-answering agent in online meetings, where it has already served more than 1,200 people.
Authors: Meng Xiao, Xunxin Cai, Qingqing Long, Chengrui Wang, Yuanchun Zhou, Hengshu Zhu
Abstract: Corpus distillation for biomedical large language models (LLMs) seeks to address the pressing challenge of insufficient quantity and quality in open-source annotated scientific corpora, which remains a bottleneck for effective LLM training in biomedical research. This paper proposes a knowledge-driven, agentic framework for scientific corpus distillation, tailored explicitly for LLM training in the biomedical domain, addressing the challenge posed by the complex hierarchy of biomedical knowledge. Central to our approach is a collaborative multi-agent architecture, where specialized agents, each guided by the Medical Subject Headings (MeSH) hierarchy, work in concert to autonomously extract, synthesize, and self-evaluate high-quality textual data from vast scientific literature. This agentic framework collectively generates and refines domain-specific question-answer pairs, ensuring comprehensive coverage and consistency with biomedical ontologies while minimizing manual involvement. Extensive experimental results show that language models trained on our multi-agent distilled datasets achieve notable improvements in biomedical question-answering tasks, outperforming both strong life sciences LLM baselines and advanced proprietary models. Notably, our AI-Ready dataset enables Llama3-70B to surpass GPT-4 with MedPrompt and Med-PaLM-2, despite their larger scale. Detailed ablation studies and case analyses further validate the effectiveness and synergy of each agent within the framework, highlighting the potential of multi-agent collaboration in biomedical LLM training.
Authors: Yanbin Wei, Xuehao Wang, Zhan Zhuang, Yang Chen, Shuhao Chen, Yulong Zhang, Yu Zhang, James Kwok
Abstract: Message-passing graph neural networks (MPNNs) and structural features (SFs) are cornerstones for the link prediction task. However, as a common and intuitive mode of understanding, the potential of visual perception has been overlooked in the MPNN community. For the first time, we equip MPNNs with vision structural awareness by proposing an effective framework called Graph Vision Network (GVN), along with a more efficient variant (E-GVN). Extensive empirical results demonstrate that with the proposed frameworks, GVN consistently benefits from the vision enhancement across seven link prediction datasets, including challenging large-scale graphs. Such improvements are compatible with existing state-of-the-art (SOTA) methods and GVNs achieve new SOTA results, thereby underscoring a promising novel direction for link prediction.
Authors: Jingren Liu, Zhong Ji, Xiangyu Chen
Abstract: While diffusion models have achieved remarkable success in static data generation, their deployment in streaming or continual learning (CL) scenarios faces a major challenge: catastrophic forgetting (CF), where newly acquired generative capabilities overwrite previously learned ones. To systematically address this, we introduce a formal Continual Diffusion Generation (CDG) paradigm that characterizes and redefines CL in the context of generative diffusion models. Prior efforts often adapt heuristic strategies from continual classification tasks but lack alignment with the underlying diffusion process. In this work, we develop the first theoretical framework for CDG by analyzing cross-task dynamics in diffusion-based generative modeling. Our analysis reveals that the retention and stability of generative knowledge across tasks are governed by three key consistency criteria: inter-task knowledge consistency (IKC), unconditional knowledge consistency (UKC), and label knowledge consistency (LKC). Building on these insights, we propose Continual Consistency Diffusion (CCD), a principled framework that integrates these consistency objectives into training via hierarchical loss terms $\mathcal{L}_{IKC}$, $\mathcal{L}_{UKC}$, and $\mathcal{L}_{LKC}$. This promotes effective knowledge retention while enabling the assimilation of new generative capabilities. Extensive experiments on four benchmark datasets demonstrate that CCD achieves state-of-the-art performance under continual settings, with substantial gains in Mean Fidelity (MF) and Incremental Mean Fidelity (IMF), particularly in tasks with rich cross-task knowledge overlap.
Authors: Yisheng Zhong, Yizhu Wen, Junfeng Guo, Mehran Kafai, Heng Huang, Hanqing Guo, Zhuangdi Zhu
Abstract: The protection of cyber Intellectual Property (IP) such as web content is an increasingly critical concern. The rise of large language models (LLMs) with online retrieval capabilities enables convenient access to information but often undermines the rights of original content creators. As users increasingly rely on LLM-generated responses, they gradually diminish direct engagement with original information sources, which will significantly reduce the incentives for IP creators to contribute, and lead to a saturating cyberspace with more AI-generated content. In response, we propose a novel defense framework that empowers web content creators to safeguard their web-based IP from unauthorized LLM real-time extraction and redistribution by leveraging the semantic understanding capability of LLMs themselves. Our method follows principled motivations and effectively addresses an intractable black-box optimization problem. Real-world experiments demonstrated that our methods improve defense success rates from 2.5% to 88.6% on different LLMs, outperforming traditional defenses such as configuration-based restrictions.
Authors: Gabriel Anzer, Kilian Arnsmeyer, Pascal Bauer, Joris Bekkers, Ulf Brefeld, Jesse Davis, Nicolas Evans, Matthias Kempe, Samuel J Robertson, Joshua Wyatt Smith, Jan Van Haaren
Abstract: During football matches, a variety of different parties (e.g., companies) each collect (possibly overlapping) data about the match ranging from basic information (e.g., starting players) to detailed positional data. This data is provided to clubs, federations, and other organizations who are increasingly interested in leveraging this data to inform their decision making. Unfortunately, analyzing such data pose significant barriers because each provider may (1) collect different data, (2) use different specifications even within the same category of data, (3) represent the data differently, and (4) delivers the data in a different manner (e.g., file format, protocol). Consequently, working with these data requires a significant investment of time and money. The goal of this work is to propose a uniform and standardized format for football data called the Common Data Format (CDF). The CDF specifies a minimal schema for five types of match data: match sheet data, video footage, event data, tracking data, and match meta data. It aims to ensure that the provided data is clear, sufficiently contextualized (e.g., its provenance is clear), and complete such that it enables common downstream analysis tasks. Concretely, this paper will detail the technical specifications of the CDF, the representational choices that were made to help ensure the clarity of the provided data, and a concrete approach for delivering data in the CDF.
Authors: Nam H. Le, Josh Bongard
Abstract: Genetic Programming (GP) has traditionally entangled the evolution of symbolic representations with their performance-based evaluation, often relying solely on raw fitness scores. This tight coupling makes GP solutions more fragile and prone to overfitting, reducing their ability to generalize. In this work, we propose LaSER (Latent Semantic Representation Regression)} -- a general framework that decouples representation evolution from lifetime learning. At each generation, candidate programs produce features which are passed to an external learner to model the target task. This approach enables any function approximator, from linear models to neural networks, to serve as a lifetime learner, allowing expressive modeling beyond conventional symbolic forms. Here we show for the first time that LaSER can outcompete standard GP and GP followed by linear regression when it employs non-linear methods to fit coefficients to GP-generated equations against complex data sets. Further, we explore how LaSER enables the emergence of innate representations, supporting long-standing hypotheses in evolutionary learning such as the Baldwin Effect. By separating the roles of representation and adaptation, LaSER offers a principled and extensible framework for symbolic regression and classification.
Authors: Hanyu Li, Haoyu Liu, Tingyu Zhu, Tianyu Guo, Zeyu Zheng, Xiaotie Deng, Michael I. Jordan
Abstract: Large Language Models (LLMs) show promise as data analysis agents, but existing benchmarks overlook the iterative nature of the field, where experts' decisions evolve with deeper insights of the dataset. To address this, we introduce IDA-Bench, a novel benchmark evaluating LLM agents in multi-round interactive scenarios. Derived from complex Kaggle notebooks, tasks are presented as sequential natural language instructions by an LLM-simulated user. Agent performance is judged by comparing its final numerical output to the human-derived baseline. Initial results show that even state-of-the-art coding agents (like Claude-3.7-thinking) succeed on < 50% of the tasks, highlighting limitations not evident in single-turn tests. This work underscores the need to improve LLMs' multi-round capabilities for building more reliable data analysis agents, highlighting the necessity of achieving a balance between instruction following and reasoning.
Authors: Charles Hong, Sahil Bhatia, Alvin Cheung, Yakun Sophia Shao
Abstract: Hardware accelerators, especially those designed for tensor processing, have become ubiquitous in today's computing landscape. However, even with significant efforts in building compilers, programming these tensor accelerators remains challenging, leaving much of their potential underutilized. Recently, large language models (LLMs), trained on large amounts of code, have shown significant promise in code generation and optimization tasks, but generating low-resource languages like specialized tensor accelerator code still poses a significant challenge. We tackle this challenge with Autocomp, an approach that empowers accelerator programmers to leverage domain knowledge and hardware feedback to optimize code via an automated LLM-driven search. We accomplish this by: 1) formulating each optimization pass as a structured two-phase prompt, divided into planning and code generation phases, 2) inserting domain knowledge during planning via a concise and adaptable optimization menu, and 3) integrating correctness and performance metrics from hardware as feedback at each search iteration. Across three categories of representative workloads and two different accelerators, we demonstrate that Autocomp-optimized code runs 5.6x (GEMM) and 2.7x (convolution) faster than the vendor-provided library, and outperforms expert-level hand-tuned code by 1.4x (GEMM), 1.1x (convolution), and 1.3x (fine-grained linear algebra). Additionally, we demonstrate that optimization schedules generated from Autocomp can be reused across similar tensor operations, improving speedups by up to 24% under a fixed sample budget.
Authors: Jintao Zhang, Xiaoming Xu, Jia Wei, Haofeng Huang, Pengle Zhang, Chendong Xiang, Jun Zhu, Jianfei Chen
Abstract: The efficiency of attention is critical because its time complexity grows quadratically with sequence length. SageAttention2 addresses this by utilizing quantization to accelerate matrix multiplications (Matmul) in attention. To further accelerate SageAttention2, we propose to utilize the faster instruction of FP8 Matmul accumulated in FP16. The instruction is 2x faster than the FP8 Matmul used in SageAttention2. Our experiments show that SageAttention2++ achieves a 3.9x speedup over FlashAttention while maintaining the same attention accuracy as SageAttention2. This means SageAttention2++ effectively accelerates various models, including those for language, image, and video generation, with negligible end-to-end metrics loss. The code will be available at https://github.com/thu-ml/SageAttention.
Authors: Henryk Mustroph, Michel Kunkler, Stefanie Rinderle-Ma
Abstract: Suffix prediction of business processes forecasts the remaining sequence of events until process completion. Current approaches focus on predicting the most likely suffix, representing a single scenario. However, when the future course of a process is subject to uncertainty and high variability, the expressiveness of such a single scenario can be limited, since other possible scenarios, which together may have a higher overall probability, are overlooked. To address this limitation, we propose probabilistic suffix prediction, a novel approach that approximates a probability distribution of suffixes. The proposed approach is based on an Uncertainty-Aware Encoder-Decoder LSTM (U-ED-LSTM) and a Monte Carlo (MC) suffix sampling algorithm. We capture epistemic uncertainties via MC dropout and aleatoric uncertainties as learned loss attenuation. This technical report presents a comprehensive evaluation of the probabilistic suffix prediction approach's predictive performance and calibration under three different hyperparameter settings, using four real-life and one artificial event log. The results show that: i) probabilistic suffix prediction can outperform most likely suffix prediction, the U-ED-LSTM has reasonable predictive performance, and ii) the model's predictions are well calibrated.
Authors: Huacan Wang, Ziyi Ni, Shuo Zhang, Shuo Lu, Sen Hu, Ziyang He, Chen Hu, Jiaye Lin, Yifu Guo, Yuntao Du, Pin Lyu
Abstract: The ultimate goal of code agents is to solve complex tasks autonomously. Although large language models (LLMs) have made substantial progress in code generation, real-world tasks typically demand full-fledged code repositories rather than simple scripts. Building such repositories from scratch remains a major challenge. Fortunately, GitHub hosts a vast, evolving collection of open-source repositories, which developers frequently reuse as modular components for complex tasks. Yet, existing frameworks like OpenHands and SWE-Agent still struggle to effectively leverage these valuable resources. Relying solely on README files provides insufficient guidance, and deeper exploration reveals two core obstacles: overwhelming information and tangled dependencies of repositories, both constrained by the limited context windows of current LLMs. To tackle these issues, we propose RepoMaster, an autonomous agent framework designed to explore and reuse GitHub repositories for solving complex tasks. For efficient understanding, RepoMaster constructs function-call graphs, module-dependency graphs, and hierarchical code trees to identify essential components, providing only identified core elements to the LLMs rather than the entire repository. During autonomous execution, it progressively explores related components using our exploration tools and prunes information to optimize context usage. Evaluated on the adjusted MLE-bench, RepoMaster achieves a 110% relative boost in valid submissions over the strongest baseline OpenHands. On our newly released GitTaskBench, RepoMaster lifts the task-pass rate from 24.1% to 62.9% while reducing token usage by 95%. Our code and demonstration materials are publicly available at https://github.com/wanghuacan/RepoMaster.
Authors: Lianghui Zhu, Xitong Ling, Minxi Ouyang, Xiaoping Liu, Tian Guan, Mingxi Fu, Zhiqiang Cheng, Fanglei Fu, Maomao Zeng, Liming Liu, Song Duan, Qiang Huang, Ying Xiao, Jianming Li, Shanming Lu, Zhenghua Piao, Mingxi Zhu, Yibo Jin, Shan Xu, Qiming He, Yizhi Wang, Junru Cheng, Xuanyu Wang, Luxi Xie, Houqiang Li, Sufang Tian, Yonghong He
Abstract: Gastrointestinal (GI) diseases represent a clinically significant burden, necessitating precise diagnostic approaches to optimize patient outcomes. Conventional histopathological diagnosis suffers from limited reproducibility and diagnostic variability. To overcome these limitations, we develop Digepath, a specialized foundation model for GI pathology. Our framework introduces a dual-phase iterative optimization strategy combining pretraining with fine-screening, specifically designed to address the detection of sparsely distributed lesion areas in whole-slide images. Digepath is pretrained on over 353 million multi-scale images from 210,043 H&E-stained slides of GI diseases. It attains state-of-the-art performance on 33 out of 34 tasks related to GI pathology, including pathological diagnosis, protein expression status prediction, gene mutation prediction, and prognosis evaluation. We further translate the intelligent screening module for early GI cancer and achieve near-perfect 99.70% sensitivity across nine independent medical institutions. This work not only advances AI-driven precision pathology for GI diseases but also bridge critical gaps in histopathological practice.
Authors: Tianjun Gu, Linfeng Li, Xuhong Wang, Chenghua Gong, Jingyu Gong, Zhizhong Zhang, Yuan Xie, Lizhuang Ma, Xin Tan
Abstract: Adaptive navigation in unfamiliar environments is crucial for household service robots but remains challenging due to the need for both low-level path planning and high-level scene understanding. While recent vision-language model (VLM) based zero-shot approaches reduce dependence on prior maps and scene-specific training data, they face significant limitations: spatiotemporal discontinuity from discrete observations, unstructured memory representations, and insufficient task understanding leading to navigation failures. We propose DORAEMON (Decentralized Ontology-aware Reliable Agent with Enhanced Memory Oriented Navigation), a novel cognitive-inspired framework consisting of Ventral and Dorsal Streams that mimics human navigation capabilities. The Dorsal Stream implements the Hierarchical Semantic-Spatial Fusion and Topology Map to handle spatiotemporal discontinuities, while the Ventral Stream combines RAG-VLM and Policy-VLM to improve decision-making. Our approach also develops Nav-Ensurance to ensure navigation safety and efficiency. We evaluate DORAEMON on the HM3D, MP3D, and GOAT datasets, where it achieves state-of-the-art performance on both success rate (SR) and success weighted by path length (SPL) metrics, significantly outperforming existing methods. We also introduce a new evaluation metric (AORI) to assess navigation intelligence better. Comprehensive experiments demonstrate DORAEMON's effectiveness in zero-shot autonomous navigation without requiring prior map building or pre-training.
Authors: Giovanny A. Cuervo-Londo\~no, Javier S\'anchez, \'Angel Rodr\'iguez-Santana
Abstract: Oceanographic forecasting impacts various sectors of society by supporting environmental conservation and economic activities. Based on global circulation models, traditional forecasting methods are computationally expensive and slow, limiting their ability to provide rapid forecasts. Recent advances in deep learning offer faster and more accurate predictions, although these data-driven models are often trained with global data from numerical simulations, which may not reflect reality. The emergence of such models presents great potential for improving ocean prediction at a subregional domain. However, their ability to predict fine-scale ocean processes, like mesoscale structures, remains largely unknown. This work aims to adapt a graph neural network initially developed for global weather forecasting to improve subregional ocean prediction, specifically focusing on the Canary Current upwelling system. The model is trained with satellite data and compared to state-of-the-art physical ocean models to assess its performance in capturing ocean dynamics. Our results show that the deep learning model surpasses traditional methods in precision despite some challenges in upwelling areas. It demonstrated superior performance in reducing RMSE errors compared to ConvLSTM and the GLORYS reanalysis, particularly in regions with complex oceanic dynamics such as Cape Ghir, Cape Bojador, and Cape Blanc. The model achieved improvements of up to 26.5% relative to ConvLSTM and error reductions of up to 76% in 5-day forecasts compared to the GLORYS reanalysis at these critical locations, highlighting its enhanced capability to capture spatial variability and improve predictive accuracy in complex areas. These findings suggest the viability of adapting meteorological data-driven models for improving subregional medium-term ocean forecasting.
Authors: Kyowoon Lee, Jaesik Choi
Abstract: Diffusion-based generative models are emerging as powerful tools for long-horizon planning in reinforcement learning (RL), particularly with offline datasets. However, their performance is fundamentally limited by the quality and diversity of training data. This often restricts their generalization to tasks outside their training distribution or longer planning horizons. To overcome this challenge, we propose State-Covering Trajectory Stitching (SCoTS), a novel reward-free trajectory augmentation method that incrementally stitches together short trajectory segments, systematically generating diverse and extended trajectories. SCoTS first learns a temporal distance-preserving latent representation that captures the underlying temporal structure of the environment, then iteratively stitches trajectory segments guided by directional exploration and novelty to effectively cover and expand this latent space. We demonstrate that SCoTS significantly improves the performance and generalization capabilities of diffusion planners on offline goal-conditioned benchmarks requiring stitching and long-horizon reasoning. Furthermore, augmented trajectories generated by SCoTS significantly improve the performance of widely used offline goal-conditioned RL algorithms across diverse environments.
Authors: Soyoung Oh, Xinting Huang, Mathis Pink, Michael Hahn, Vera Demberg
Abstract: Idioms present a unique challenge for language models due to their non-compositional figurative meanings, which often strongly diverge from the idiom's literal interpretation. This duality requires a model to learn representing and deciding between the two meanings to interpret an idiom in a figurative sense, or literally. In this paper, we employ tools from mechanistic interpretability to trace how a large pretrained causal transformer (LLama3.2-1B-base) deals with this ambiguity. We localize three steps of idiom processing: First, the idiom's figurative meaning is retrieved in early attention and MLP sublayers. We identify specific attention heads which boost the figurative meaning of the idiom while suppressing the idiom's literal interpretation. The model subsequently represents the figurative representation through an intermediate path. Meanwhile, a parallel bypass route forwards literal interpretation, ensuring that a both reading remain available. Overall, our findings provide a mechanistic evidence for idiom comprehension in an autoregressive transformer.
Authors: Adrian de Wynter
Abstract: An evaluator is trustworthy when there exists some agreed-upon way to measure its performance as a labeller. The two ways to establish trustworthiness are either by testing it, or by assuming the evaluator `knows' somehow the way to label the corpus. However, if labelled references (e.g., a development set) are unavailable, neither of these approaches work: the former requires the data, and the latter is an assumption, not evidence. To address this, we introduce an algorithm (the `No-Data Algorithm') by which to establish trust in an evaluator without any existing references. Our algorithm works by successively posing challenges to said evaluator. We show that this is sufficient to establish trustworthiness w.h.p., in such a way that when the evaluator actually knows the way to label the corpus, the No-Data Algorithm accepts its output; and, conversely, flags untrustworthy evaluators when these are unable to prove it. We present formal proofs of correctness and limited experiments.
Authors: Rui Yann, Xianglei Xing
Abstract: We present SemiOccam, an image recognition network that leverages semi-supervised learning in a highly efficient manner. Existing works often rely on complex training techniques and architectures, requiring hundreds of GPU hours for training, while their generalization ability when dealing with extremely limited labeled data remains to be improved. To address these limitations, we construct a hierarchical mixture density classification decision mechanism by optimizing mutual information between feature representations and target classes, compressing redundant information while retaining crucial discriminative components. Experimental results demonstrate that our method achieves state-of-the-art performance on various datasets when using negligible labeled samples, and its simple architecture keeps training time to minute-level. Notably, this paper reveals a long-overlooked data leakage issue in the STL-10 dataset for semi-supervised learning tasks and removes duplicates to ensure the reliability of experimental results. We also release the deduplicated CleanSTL-10 dataset to facilitate fair and reliable research in future semi-supervised learning. Code available at https://github.com/Shu1L0n9/SemiOccam.
Authors: Yanting Wang, Wei Zou, Runpeng Geng, Jinyuan Jia
Abstract: Long context large language models (LLMs) are deployed in many real-world applications such as RAG, agent, and broad LLM-integrated applications. Given an instruction and a long context (e.g., documents, PDF files, webpages), a long context LLM can generate an output grounded in the provided context, aiming to provide more accurate, up-to-date, and verifiable outputs while reducing hallucinations and unsupported claims. This raises a research question: how to pinpoint the texts (e.g., sentences, passages, or paragraphs) in the context that contribute most to or are responsible for the generated output by an LLM? This process, which we call context traceback, has various real-world applications, such as 1) debugging LLM-based systems, 2) conducting post-attack forensic analysis for attacks (e.g., prompt injection attack, knowledge corruption attacks) to an LLM, and 3) highlighting knowledge sources to enhance the trust of users towards outputs generated by LLMs. When applied to context traceback for long context LLMs, existing feature attribution methods such as Shapley have sub-optimal performance and/or incur a large computational cost. In this work, we develop TracLLM, the first generic context traceback framework tailored to long context LLMs. Our framework can improve the effectiveness and efficiency of existing feature attribution methods. To improve the efficiency, we develop an informed search based algorithm in TracLLM. We also develop contribution score ensemble/denoising techniques to improve the accuracy of TracLLM. Our evaluation results show TracLLM can effectively identify texts in a long context that lead to the output of an LLM. Our code and data are at: https://github.com/Wang-Yanting/TracLLM.
Authors: Takayuki Komatsu, Yoshiyuki Ohmura, Kayato Nishitsunoi, Yasuo Kuniyoshi
Abstract: The main goal of representation learning is to acquire meaningful representations from real-world sensory inputs without supervision. Representation learning explains some aspects of human development. Various neural network (NN) models have been proposed that acquire empirically good representations. However, the formulation of a good representation has not been established. We recently proposed a method for categorizing changes between a pair of sensory inputs. A unique feature of this approach is that transformations between two sensory inputs are learned to satisfy algebraic structural constraints. Conventional representation learning often assumes that disentangled independent feature axes is a good representation; however, we found that such a representation cannot account for conditional independence. To overcome this problem, we proposed a new method using group decomposition in Galois algebra theory. Although this method is promising for defining a more general representation, it assumes pixel-to-pixel translation without feature extraction, and can only process low-resolution images with no background, which prevents real-world application. In this study, we provide a simple method to apply our group decomposition theory to a more realistic scenario by combining feature extraction and object segmentation. We replace pixel translation with feature translation and formulate object segmentation as grouping features under the same transformation. We validated the proposed method on a practical dataset containing both real-world object and background. We believe that our model will lead to a better understanding of human development of object recognition in the real world.
Authors: Yin Lu, Xuening Zhu, Tong He, David Wipf
Abstract: Is there really much more to say about sparse autoencoders (SAEs)? Autoencoders in general, and SAEs in particular, represent deep architectures that are capable of modeling low-dimensional latent structure in data. Such structure could reflect, among other things, correlation patterns in large language model activations, or complex natural image manifolds. And yet despite the wide-ranging applicability, there have been relatively few changes to SAEs beyond the original recipe from decades ago, namely, standard deep encoder/decoder layers trained with a classical/deterministic sparse regularizer applied within the latent space. One possible exception is the variational autoencoder (VAE), which adopts a stochastic encoder module capable of producing sparse representations when applied to manifold data. In this work we formalize underappreciated weaknesses with both canonical SAEs, as well as analogous VAEs applied to similar tasks, and propose a hybrid alternative model that circumvents these prior limitations. In terms of theoretical support, we prove that global minima of our proposed model recover certain forms of structured data spread across a union of manifolds. Meanwhile, empirical evaluations on synthetic and real-world datasets substantiate the efficacy of our approach in accurately estimating underlying manifold dimensions and producing sparser latent representations without compromising reconstruction error. In general, we are able to exceed the performance of equivalent-capacity SAEs and VAEs, as well as recent diffusion models where applicable, within domains such as images and language model activation patterns.
Authors: Iulia M. Comsa, Murray Shanahan
Abstract: Large language models (LLMs) exhibit compelling linguistic behaviour, and sometimes offer self-reports, that is to say statements about their own nature, inner workings, or behaviour. In humans, such reports are often attributed to a faculty of introspection and are typically linked to consciousness. This raises the question of how to interpret self-reports produced by LLMs, given their increasing linguistic fluency and cognitive capabilities. To what extent (if any) can the concept of introspection be meaningfully applied to LLMs? Here, we present and critique two examples of apparent introspective self-report from LLMs. In the first example, an LLM attempts to describe the process behind its own "creative" writing, and we argue this is not a valid example of introspection. In the second example, an LLM correctly infers the value of its own temperature parameter, and we argue that this can be legitimately considered a minimal example of introspection, albeit one that is (presumably) not accompanied by conscious experience.
Authors: Bhavik Chandna, Zubair Bashir, Procheta Sen
Abstract: Large Language Models (LLMs) are known to exhibit social, demographic, and gender biases, often as a consequence of the data on which they are trained. In this work, we adopt a mechanistic interpretability approach to analyze how such biases are structurally represented within models such as GPT-2 and Llama2. Focusing on demographic and gender biases, we explore different metrics to identify the internal edges responsible for biased behavior. We then assess the stability, localization, and generalizability of these components across dataset and linguistic variations. Through systematic ablations, we demonstrate that bias-related computations are highly localized, often concentrated in a small subset of layers. Moreover, the identified components change across fine-tuning settings, including those unrelated to bias. Finally, we show that removing these components not only reduces biased outputs but also affects other NLP tasks, such as named entity recognition and linguistic acceptability judgment because of the sharing of important components with these tasks.
Authors: Yeonseok Jeong, Jinsu Kim, Dohyeon Lee, Seung-won Hwang
Abstract: Large Language Models (LLMs) have shown remarkable performance in Open-Domain Question Answering (ODQA) by leveraging external documents through Retrieval-Augmented Generation (RAG). To reduce RAG overhead, from longer context, context compression is necessary. However, prior compression methods do not focus on filtering out non-evidential information, which limit the performance in LLM-based RAG. We thus propose Evidentiality-guided RAG, or ECoRAG framework. ECoRAG improves LLM performance by compressing retrieved documents based on evidentiality, ensuring whether answer generation is supported by the correct evidence. As an additional step, ECoRAG reflects whether the compressed content provides sufficient evidence, and if not, retrieves more until sufficient. Experiments show that ECoRAG improves LLM performance on ODQA tasks, outperforming existing compression methods. Furthermore, ECoRAG is highly cost-efficient, as it not only reduces latency but also minimizes token usage by retaining only the necessary information to generate the correct answer. Code is available at https://github.com/ldilab/ECoRAG.
Authors: Mohammed Almutairi
Abstract: Effective teamwork is essential across diverse domains. During the team formation stage, a key challenge is forming teams that effectively balance user preferences with task objectives to enhance overall team satisfaction. In the team performing stage, maintaining cohesion and engagement is critical for sustaining high team performance. However, existing computational tools and algorithms for team optimization often rely on static data inputs, narrow algorithmic objectives, or solutions tailored for specific contexts, failing to account for the dynamic interplay of team members personalities, evolving goals, and changing individual preferences. Therefore, teams may encounter member dissatisfaction, as purely algorithmic assignments can reduce members commitment to team goals or experience suboptimal engagement due to the absence of timely, personalized guidance to help members adjust their behaviors and interactions as team dynamics evolve. Ultimately, these challenges can lead to reduced overall team performance. My Ph.D. dissertation aims to develop AI-augmented team optimization frameworks and practical systems that enhance team satisfaction, engagement, and performance. First, I propose a team formation framework that leverages a multi-armed bandit algorithm to iteratively refine team composition based on user preferences, ensuring alignment between individual needs and collective team goals to enhance team satisfaction. Second, I introduce tAIfa (Team AI Feedback Assistant), an AI-powered system that utilizes large language models (LLMs) to deliver immediate, personalized feedback to both teams and individual members, enhancing cohesion and engagement. Finally, I present PuppeteerLLM, an LLM-based simulation framework that simulates multi-agent teams to model complex team dynamics within realistic environments, incorporating task-driven collaboration and long-term coordination.
Authors: Keshigeyan Chandrasegaran, Michael Poli, Daniel Y. Fu, Dongjun Kim, Lea M. Hadzic, Manling Li, Agrim Gupta, Stefano Massaroli, Azalia Mirhoseini, Juan Carlos Niebles, Stefano Ermon, Li Fei-Fei
Abstract: Designing model architectures requires decisions such as selecting operators (e.g., attention, convolution) and configurations (e.g., depth, width). However, evaluating the impact of these decisions on model quality requires costly pretraining, limiting architectural investigation. Inspired by how new software is built on existing code, we ask: can new architecture designs be studied using pretrained models? To this end, we present grafting, a simple approach for editing pretrained diffusion transformers (DiTs) to materialize new architectures under small compute budgets. Informed by our analysis of activation behavior and attention locality, we construct a testbed based on the DiT-XL/2 design to study the impact of grafting on model quality. Using this testbed, we develop a family of hybrid designs via grafting: replacing softmax attention with gated convolution, local attention, and linear attention, and replacing MLPs with variable expansion ratio and convolutional variants. Notably, many hybrid designs achieve good quality (FID: 2.38-2.64 vs. 2.27 for DiT-XL/2) using <2% pretraining compute. We then graft a text-to-image model (PixArt-Sigma), achieving a 1.43x speedup with less than a 2% drop in GenEval score. Finally, we present a case study that restructures DiT-XL/2 by converting every pair of sequential transformer blocks into parallel blocks via grafting. This reduces model depth by 2x and yields better quality (FID: 2.77) than other models of comparable depth. Together, we show that new diffusion model designs can be explored by grafting pretrained DiTs, with edits ranging from operator replacement to architecture restructuring. Code and grafted models: https://grafting.stanford.edu