Authors: Junyu Liu, Siwen Yang, Dexiu Ma, Qian Niu, Zequn Zhang, Momoko Nagai-Tanima, Tomoki Aoyama
Abstract: Human papillomavirus (HPV) vaccine hesitancy poses significant public health challenges, particularly in Japan where proactive vaccination recommendations were suspended from 2013 to 2021. The resulting information gap is exacerbated by misinformation on social media, and traditional ways cannot simultaneously address individual queries while monitoring population-level discourse. This study aimed to develop a dual-purpose AI agent system that provides verified HPV vaccine information through a conversational interface while generating analytical reports for medical institutions based on user interactions and social media. We implemented a system comprising: a vector database integrating academic papers, government sources, news media, and social media; a Retrieval-Augmented Generation chatbot using ReAct agent architecture with multi-tool orchestration across five knowledge sources; and an automated report generation system with modules for news analysis, research synthesis, social media sentiment analysis, and user interaction pattern identification. Performance was assessed using a 0-5 scoring scale. For single-turn evaluation, the chatbot achieved mean scores of 4.83 for relevance, 4.89 for routing, 4.50 for reference quality, 4.90 for correctness, and 4.88 for professional identity (overall 4.80). Multi-turn evaluation yielded higher scores: context retention 4.94, topic coherence 5.00, and overall 4.98. The report generation system achieved completeness 4.00-5.00, correctness 4.00-5.00, and helpfulness 3.67-5.00, with reference validity 5.00 across all periods. This study demonstrates the feasibility of an integrated AI agent system for bidirectional HPV vaccine communication. The architecture enables verified information delivery with source attribution while providing systematic public discourse analysis, with a transferable framework for adaptation to other medical contexts.
Authors: Gerard Yeo, Svetlana Churina, Kokil Jaidka
Abstract: Perceived trustworthiness underpins how users navigate online information, yet it remains unclear whether large language models (LLMs),increasingly embedded in search, recommendation, and conversational systems, represent this construct in psychologically coherent ways. We analyze how instruction-tuned LLMs (Llama 3.1 8B, Qwen 2.5 7B, Mistral 7B) encode perceived trustworthiness in web-like narratives using the PEACE-Reviews dataset annotated for cognitive appraisals, emotions, and behavioral intentions. Across models, systematic layer- and head-level activation differences distinguish high- from low-trust texts, revealing that trust cues are implicitly encoded during pretraining. Probing analyses show linearly de-codable trust signals and fine-tuning effects that refine rather than restructure these representations. Strongest associations emerge with appraisals of fairness, certainty, and accountability-self -- dimensions central to human trust formation online. These findings demonstrate that modern LLMs internalize psychologically grounded trust signals without explicit supervision, offering a representational foundation for designing credible, transparent, and trust-worthy AI systems in the web ecosystem. Code and appendix are available at: https://github.com/GerardYeo/TrustworthinessLLM.
Authors: Ross Chu, Yuting Huang
Abstract: This paper develops AI agents that help job seekers write effective requests for job referrals in a professional online community. The basic workflow consists of an improver agent that rewrites the referral request and an evaluator agent that measures the quality of revisions using a model trained to predict the probability of receiving referrals from other users. Revisions suggested by the LLM (large language model) increase predicted success rates for weaker requests while reducing them for stronger requests. Enhancing the LLM with Retrieval-Augmented Generation (RAG) prevents edits that worsen stronger requests while it amplifies improvements for weaker requests. Overall, using LLM revisions with RAG increases the predicted success rate for weaker requests by 14\% without degrading performance on stronger requests. Although improvements in model-predicted success do not guarantee more referrals in the real world, they provide low-cost signals for promising features before running higher-stakes experiments on real users.
Authors: Xinyue Ma, Heelim Hong, Taegeon Um, Jongseop Lee, Seoyeong Choy, Woo-Yeon Lee, Myeongjae Jeon
Abstract: Serving long-context LLMs is challenging because request lengths and batch composition vary during token generation, causing the memory footprint to fluctuate significantly at runtime. Offloading KV caches to host memory limits effective memory usage, but existing static and predetermined offloading strategies cannot adapt to the rapidly shifting memory demands of long-context serving. This often leads to excessive CPU-to-GPU KV transfers that translate into latency spikes and frequent SLO violations. To address these challenges, we introduce ORBITFLOW, a fine-grained and adaptive KV cache management system that meets latency SLOs in long-context LLM serving. ORBITFLOW employs a lightweight ILP solver to decide which layers' KV caches to retain on the GPU for each request, within memory capacity constraints. It continuously refines KV placements based on runtime feedback when the active plan becomes suboptimal during token generation. Under heavy load, ORBITFLOW invokes a fallback mechanism to temporarily defer in-flight requests with large memory footprints, preserving overall SLO attainment. Our experiments demonstrate that ORBITFLOW improves SLO attainment for TPOT and TBT by up to 66% and 48%, respectively, while reducing the 95th percentile latency by 38% and achieving up to 3.3x higher throughput compared to existing offloading methods.
Authors: Percy Jardine
Abstract: Recently, multi-time-scale agent architectures have extended the ubiquitous single-loop paradigm by introducing temporal hierarchies with distinct cognitive layers. While yielding substantial performance gains, this diversification fundamentally compromises the coordination stability intrinsic to unified agent systems, which causes severe inter-layer conflicts, unbounded error propagation, and restricted scalability. To address these challenges, we propose Constrained Temporal Hierarchical Architecture (CTHA), a general framework that projects the inter-layer communication space onto structured manifolds to restore coordination stability, while incorporating principled arbitration mechanisms to ensure coherent decision-making. Specifically, CTHA enforces three key constraints: (1) Message Contract Constraints that formalize information flow between layers via typed summary, plan, and policy packets; (2) Authority Manifold Constraints that bound each layer's decision space according to its temporal scope; and (3) Arbiter Resolution Constraints that guarantee conflict-free composition of multi-layer decisions. Empirical experiments demonstrate that CTHA is effective for complex task execution at scale, offering 47% reduction in failure cascades, 2.3x improvement in sample efficiency, and superior scalability compared to unconstrained hierarchical baselines. We anticipate that CTHA, as a principled extension of temporal hierarchies, will contribute to a deeper understanding of multi-agent coordination and suggest promising directions for the evolution of robust autonomous systems.
Authors: Sen Wang, Bangwei Liu, Zhenkun Gao, Lizhuang Ma, Xuhong Wang, Yuan Xie, Xin Tan
Abstract: An ideal embodied agent should possess lifelong learning capabilities to handle long-horizon and complex tasks, enabling continuous operation in general environments. This not only requires the agent to accurately accomplish given tasks but also to leverage long-term episodic memory to optimize decision-making. However, existing mainstream one-shot embodied tasks primarily focus on task completion results, neglecting the crucial process of exploration and memory utilization. To address this, we propose Long-term Memory Embodied Exploration (LMEE), which aims to unify the agent's exploratory cognition and decision-making behaviors to promote lifelong learning.We further construct a corresponding dataset and benchmark, LMEE-Bench, incorporating multi-goal navigation and memory-based question answering to comprehensively evaluate both the process and outcome of embodied exploration. To enhance the agent's memory recall and proactive exploration capabilities, we propose MemoryExplorer, a novel method that fine-tunes a multimodal large language model through reinforcement learning to encourage active memory querying. By incorporating a multi-task reward function that includes action prediction, frontier selection, and question answering, our model achieves proactive exploration. Extensive experiments against state-of-the-art embodied exploration models demonstrate that our approach achieves significant advantages in long-horizon embodied tasks.
Authors: Nina Bo\v{c}kov\'a, Barbora Voln\'a, Mirko Dohnal
Abstract: This paper investigates complex product-innovation processes using models grounded in a set of heuristics. Each heuristic is expressed through simple trends -- increasing, decreasing, or constant -- which serve as minimally information-intensive quantifiers, avoiding reliance on numerical values or rough sets. A solution to a trend model is defined as a set of scenarios with possible transitions between them, represented by a transition graph. Any possible future or past behaviour of the system under study can thus be depicted by a path within this graph.
Authors: Fran\c{c}ois Chollet, Mike Knoop, Gregory Kamradt, Bryan Landers
Abstract: The ARC-AGI benchmark series serves as a critical measure of few-shot generalization on novel tasks, a core aspect of intelligence. The ARC Prize 2025 global competition targeted the newly released ARC-AGI-2 dataset, which features greater task complexity compared to its predecessor. The Kaggle competition attracted 1,455 teams and 15,154 entries, with the top score reaching 24% on the ARC-AGI-2 private evaluation set. Paper submissions nearly doubled year-over-year to 90 entries, reflecting the growing research interest in fluid intelligence and abstract reasoning. The defining theme of 2025 is the emergence of the refinement loop -- a per-task iterative program optimization loop guided by a feedback signal. Refinement loops come in a variety of forms, in particular evolutionary program synthesis approaches and application-layer refinements to commercial AI systems. Such refinement loops are also possible in weight space, as evidenced by zero-pretraining deep learning methods which are now achieving competitive performance with remarkably small networks (7M parameters). In parallel, four frontier AI labs (Anthropic, Google DeepMind, OpenAI, and xAI) reported ARC-AGI performance in public model cards in 2025, establishing ARC-AGI as an industry standard benchmark for AI reasoning. However, our analysis indicates that current frontier AI reasoning performance remains fundamentally constrained to knowledge coverage, giving rise to new forms of benchmark contamination. In this paper, we survey the top-performing methods, examine the role of refinement loops in AGI progress, discuss knowledge-dependent overfitting, and preview ARC-AGI-3, which introduces interactive reasoning challenges that require exploration, planning, memory, goal acquisition, and alignment capabilities.
Authors: Yosub Shin, Michael Buriek, Boris Sobolev, Pavel Bushuyeu, Vikas Kumar, Haoyang Xu, Samuel Watson, Igor Molybog
Abstract: We study data curation for multimodal reasoning through the NeurIPS 2025 Data Curation for Vision-Language Reasoning (DCVLR) challenge, which isolates dataset selection by fixing the model and training protocol. Using a compact curated dataset derived primarily from Walton Multimodal Cold Start, our submission placed first in the challenge. Through post-competition ablations, we show that difficulty-based example selection on an aligned base dataset is the dominant driver of performance gains. Increasing dataset size does not reliably improve mean accuracy under the fixed training recipe, but mainly reduces run-to-run variance, while commonly used diversity and synthetic augmentation heuristics provide no additional benefit and often degrade performance. These results characterize DCVLR as a saturation-regime evaluation and highlight the central role of alignment and difficulty in data-efficient multimodal reasoning.
Authors: Zhenhua Xu, Dongsheng Chen, Shuo Wang, Jian Li, Chengjie Wang, Meng Han, Yabiao Wang
Abstract: LLM role-playing aims to portray arbitrary characters in interactive narratives, yet existing systems often suffer from limited immersion and adaptability. They typically under-model dynamic environmental information and assume largely static scenes and casts, offering insufficient support for multi-character orchestration, scene transitions, and on-the-fly character introduction. We propose an adaptive multi-agent role-playing framework, AdaMARP, featuring an immersive message format that interleaves [Thought], (Action),
Authors: Jiahao Wang, Shuangjia Zheng
Abstract: The ability to engineer optimized protein variants has transformative potential for biotechnology and medicine. Prior sequence-based optimization methods struggle with the high-dimensional complexities due to the epistasis effect and the disregard for structural constraints. To address this, we propose HADES, a Bayesian optimization method utilizing Hamiltonian dynamics to efficiently sample from a structure-aware approximated posterior. Leveraging momentum and uncertainty in the simulated physical movements, HADES enables rapid transition of proposals toward promising areas. A position discretization procedure is introduced to propose discrete protein sequences from such a continuous state system. The posterior surrogate is powered by a two-stage encoder-decoder framework to determine the structure and function relationships between mutant neighbors, consequently learning a smoothed landscape to sample from. Extensive experiments demonstrate that our method outperforms state-of-the-art baselines in in-silico evaluations across most metrics. Remarkably, our approach offers a unique advantage by leveraging the mutual constraints between protein structure and sequence, facilitating the design of protein sequences with similar structures and optimized properties. The code and data are publicly available at https://github.com/GENTEL-lab/HADES.
Authors: Shiyu Liu, Yongjing Yin, Jianhao Yan, Yunbo Tang, Qinggang Zhang, Bei Li, Xin Chen, Jingang Wang, Xunliang Cai, Jinsong Su
Abstract: RL-based agentic search enables LLMs to solve complex questions via dynamic planning and external search. While this approach significantly enhances accuracy with agent policies optimized via large-scale reinforcement learning, we identify a critical gap in reliability: these agents fail to recognize their reasoning boundaries and rarely admit ``I DON'T KNOW'' (IDK) even when evidence is insufficient or reasoning reaches its limit. The lack of reliability often leads to plausible but unreliable answers, introducing significant risks in many real-world scenarios. To this end, we propose Boundary-Aware Policy Optimization (BAPO), a novel RL framework designed to cultivate reliable boundary awareness without compromising accuracy. BAPO introduces two key components: (i) a group-based boundary-aware reward that encourages an IDK response only when the reasoning reaches its limit, and (ii) an adaptive reward modulator that strategically suspends this reward during early exploration, preventing the model from exploiting IDK as a shortcut. Extensive experiments on four benchmarks demonstrate that BAPO substantially enhances the overall reliability of agentic search.
Authors: Keyu Li, Junhao Shi, Yang Xiao, Mohan Jiang, Jie Sun, Yunze Wu, Shijie Xia, Xiaojie Cai, Tianze Xu, Weiye Si, Wenjie Li, Dequan Wang, Pengfei Liu
Abstract: Large Language Models (LLMs) based autonomous agents demonstrate multifaceted capabilities to contribute substantially to economic production. However, existing benchmarks remain focused on single agentic capability, failing to capture long-horizon real-world scenarios. Moreover, the reliance on human-in-the-loop feedback for realistic tasks creates a scalability bottleneck, hindering automated rollout collection and evaluation. To bridge this gap, we introduce AgencyBench, a comprehensive benchmark derived from daily AI usage, evaluating 6 core agentic capabilities across 32 real-world scenarios, comprising 138 tasks with specific queries, deliverables, and rubrics. These scenarios require an average of 90 tool calls, 1 million tokens, and hours of execution time to resolve. To enable automated evaluation, we employ a user simulation agent to provide iterative feedback, and a Docker sandbox to conduct visual and functional rubric-based assessment. Experiments reveal that closed-source models significantly outperform open-source models (48.4% vs 32.1%). Further analysis reveals significant disparities across models in resource efficiency, feedback-driven self-correction, and specific tool-use preferences. Finally, we investigate the impact of agentic scaffolds, observing that proprietary models demonstrate superior performance within their native ecosystems (e.g., Claude-4.5-Opus via Claude-Agent-SDK), while open-source models exhibit distinct performance peaks, suggesting potential optimization for specific execution frameworks. AgencyBench serves as a critical testbed for next-generation agents, highlighting the necessity of co-optimizing model architecture with agentic frameworks. We believe this work sheds light on the future direction of autonomous agents, and we release the full benchmark and evaluation toolkit at https://github.com/GAIR-NLP/AgencyBench.
Authors: Suhan Guo, Jiahong Deng, Furao Shen
Abstract: Accurate forecasting of infectious disease dynamics is critical for public health planning and intervention. Human mobility plays a central role in shaping the spatial spread of epidemics, but mobility data are noisy, indirect, and difficult to integrate reliably with disease records. Meanwhile, epidemic case time series are typically short and reported at coarse temporal resolution. These conditions limit the effectiveness of parameter-heavy mobility-aware forecasters that rely on clean and abundant data. In this work, we propose the Mobility-Informed Causal Adapter (MiCA), a lightweight and architecture-agnostic module for epidemic forecasting. MiCA infers mobility relations through causal discovery and integrates them into temporal forecasting models via gated residual mixing. This design allows lightweight forecasters to selectively exploit mobility-derived spatial structure while remaining robust under noisy and data-limited conditions, without introducing heavy relational components such as graph neural networks or full attention. Extensive experiments on four real-world epidemic datasets, including COVID-19 incidence, COVID-19 mortality, influenza, and dengue, show that MiCA consistently improves lightweight temporal backbones, achieving an average relative error reduction of 7.5\% across forecasting horizons. Moreover, MiCA attains performance competitive with SOTA spatio-temporal models while remaining lightweight.
Authors: Zhezheng Hao, Hong Wang, Jian Luo, Jianqing Zhang, Yuyan Zhou, Qiang Lin, Can Wang, Hande Dong, Jiawei Chen
Abstract: Large Language Model agents are reshaping the industrial landscape. However, most practical agents remain human-designed because tasks differ widely, making them labor-intensive to build. This situation poses a central question: can we automatically create and adapt domain agents in the wild? While several recent approaches have sought to automate agent creation, they typically treat agent generation as a black-box procedure and rely solely on final performance metrics to guide the process. Such strategies overlook critical evidence explaining why an agent succeeds or fails, and often require high computational costs. To address these limitations, we propose ReCreate, an experience-driven framework for the automatic creation of domain agents. ReCreate systematically leverages agent interaction histories, which provide rich concrete signals on both the causes of success or failure and the avenues for improvement. Specifically, we introduce an agent-as-optimizer paradigm that effectively learns from experience via three key components: (i) an experience storage and retrieval mechanism for on-demand inspection; (ii) a reasoning-creating synergy pipeline that maps execution experience into scaffold edits; and (iii) hierarchical updates that abstract instance-level details into reusable domain patterns. In experiments across diverse domains, ReCreate consistently outperforms human-designed agents and existing automated agent generation methods, even when starting from minimal seed scaffolds.
Authors: Zixu Wang, Bingbing Xu, Yige Yuan, Huawei Shen, Xueqi Cheng
Abstract: Multi-Agent Systems (MAS) built on large language models typically solve complex tasks by coordinating multiple agents through workflows. Existing approaches generates workflows either at task level or query level, but their relative costs and benefits remain unclear. After rethinking and empirical analyses, we show that query-level workflow generation is not always necessary, since a small set of top-K best task-level workflows together already covers equivalent or even more queries. We further find that exhaustive execution-based task-level evaluation is both extremely token-costly and frequently unreliable. Inspired by the idea of self-evolution and generative reward modeling, we propose a low-cost task-level generation framework \textbf{SCALE}, which means \underline{\textbf{S}}elf prediction of the optimizer with few shot \underline{\textbf{CAL}}ibration for \underline{\textbf{E}}valuation instead of full validation execution. Extensive experiments demonstrate that \textbf{SCALE} maintains competitive performance, with an average degradation of just 0.61\% compared to existing approach across multiple datasets, while cutting overall token usage by up to 83\%.
Authors: Girish A. Koushik, Helen Treharne, Diptesh Kanojia
Abstract: Social media platforms are increasingly dominated by long-form multimodal content, where harmful narratives are constructed through a complex interplay of audio, visual, and textual cues. While automated systems can flag hate speech with high accuracy, they often function as "black boxes" that fail to provide the granular, interpretable evidence, such as precise timestamps and target identities, required for effective human-in-the-loop moderation. In this work, we introduce TANDEM, a unified framework that transforms audio-visual hate detection from a binary classification task into a structured reasoning problem. Our approach employs a novel tandem reinforcement learning strategy where vision-language and audio-language models optimize each other through self-constrained cross-modal context, stabilizing reasoning over extended temporal sequences without requiring dense frame-level supervision. Experiments across three benchmark datasets demonstrate that TANDEM significantly outperforms zero-shot and context-augmented baselines, achieving 0.73 F1 in target identification on HateMM (a 30% improvement over state-of-the-art) while maintaining precise temporal grounding. We further observe that while binary detection is robust, differentiating between offensive and hateful content remains challenging in multi-class settings due to inherent label ambiguity and dataset imbalance. More broadly, our findings suggest that structured, interpretable alignment is achievable even in complex multimodal settings, offering a blueprint for the next generation of transparent and actionable online safety moderation tools.
Authors: Sofiene Lassoued, Asrat Gobachew, Stefan Lier, Andreas Schwung
Abstract: This paper proposes a policy-based deep reinforcement learning hyper-heuristic framework for solving the Job Shop Scheduling Problem. The hyper-heuristic agent learns to switch scheduling rules based on the system state dynamically. We extend the hyper-heuristic framework with two key mechanisms. First, action prefiltering restricts decision-making to feasible low-level actions, enabling low-level heuristics to be evaluated independently of environmental constraints and providing an unbiased assessment. Second, a commitment mechanism regulates the frequency of heuristic switching. We investigate the impact of different commitment strategies, from step-wise switching to full-episode commitment, on both training behavior and makespan. Additionally, we compare two action selection strategies at the policy level: deterministic greedy selection and stochastic sampling. Computational experiments on standard JSSP benchmarks demonstrate that the proposed approach outperforms traditional heuristics, metaheuristics, and recent neural network-based scheduling methods
Authors: Qianyue Wang, Jinwu Hu, Yufeng Wang, Huanxiang Lin, Bolin Chen, Zhiquan Wen, Yaofo Chen, Mingkui Tan
Abstract: Large Reasoning Models (LRMs) excel at multi-step reasoning but often suffer from inefficient reasoning processes like overthinking and overshoot, where excessive or misdirected reasoning increases computational cost and degrades performance. Existing efficient reasoning methods operate in a closed-loop manner, lacking mechanisms for external intervention to guide the reasoning process. To address this, we propose Think-with-Me, a novel test-time interactive reasoning paradigm that introduces external feedback intervention into the reasoning process. Our key insights are that transitional conjunctions serve as natural points for intervention, signaling phases of self-validation or exploration and using transitional words appropriately to prolong the reasoning enhances performance, while excessive use affects performance. Building on these insights, Think-with-Me pauses reasoning at these points for external feedback, adaptively extending or terminating reasoning to reduce redundancy while preserving accuracy. The feedback is generated via a multi-criteria evaluation (rationality and completeness) and comes from either human or LLM proxies. We train the target model using Group Relative Policy Optimization (GRPO) to adapt to this interactive mode. Experiments show that Think-with-Me achieves a superior balance between accuracy and reasoning length under limited context windows. On AIME24, Think-with-Me outperforms QwQ-32B by 7.19% in accuracy while reducing average reasoning length by 81% under an 8K window. The paradigm also benefits security and creative tasks.
Authors: Weihong Qi, Fan Huang, Rasika Muralidharan, Jisun An, Haewoon Kwak
Abstract: We present XChoice, an explainable framework for evaluating AI-human alignment in constrained decision making. Moving beyond outcome agreement such as accuracy and F1 score, XChoice fits a mechanism-based decision model to human data and LLM-generated decisions, recovering interpretable parameters that capture the relative importance of decision factors, constraint sensitivity, and implied trade-offs. Alignment is assessed by comparing these parameter vectors across models, options, and subgroups. We demonstrate XChoice on Americans' daily time allocation using the American Time Use Survey (ATUS) as human ground truth, revealing heterogeneous alignment across models and activities and salient misalignment concentrated in Black and married groups. We further validate robustness of XChoice via an invariance analysis and evaluate targeted mitigation with a retrieval augmented generation (RAG) intervention. Overall, XChoice provides mechanism-based metrics that diagnose misalignment and support informed improvements beyond surface outcome matching.
Authors: Weiyi Wang, Xinchi Chen, Jingjing Gong, Xuanjing Huang, Xipeng Qiu
Abstract: Recent advances in agentic Large Language Models (LLMs) have positioned them as generalist planners capable of reasoning and acting across diverse tasks. However, existing agent benchmarks largely focus on symbolic or weakly grounded environments, leaving their performance in physics-constrained real-world domains underexplored. We introduce AstroReason-Bench, a comprehensive benchmark for evaluating agentic planning in Space Planning Problems (SPP), a family of high-stakes problems with heterogeneous objectives, strict physical constraints, and long-horizon decision-making. AstroReason-Bench integrates multiple scheduling regimes, including ground station communication and agile Earth observation, and provides a unified agent-oriented interaction protocol. Evaluating on a range of state-of-the-art open- and closed-source agentic LLM systems, we find that current agents substantially underperform specialized solvers, highlighting key limitations of generalist planning under realistic constraints. AstroReason-Bench offers a challenging and diagnostic testbed for future agentic research.
Authors: Hedieh Haddad, Thibault Falque, Pierre Talbot, Pascal Bouvry
Abstract: The performance of constraint programming solvers is highly sensitive to the choice of their hyperparameters. Manually finding the best solver configuration is a difficult, time-consuming task that typically requires expert knowledge. In this paper, we introduce probe and solve algorithm, a novel two-phase framework for automated hyperparameter optimization integrated into the CPMpy library. This approach partitions the available time budget into two phases: a probing phase that explores different sets of hyperparameters using configurable hyperparameter optimization methods, followed by a solving phase where the best configuration found is used to tackle the problem within the remaining time. We implement and compare two hyperparameter optimization methods within the probe and solve algorithm: Bayesian optimization and Hamming distance search. We evaluate the algorithm on two different constraint programming solvers, ACE and Choco, across 114 combinatorial problem instances, comparing their performance against the solver's default configurations. Results show that using Bayesian optimization, the algorithm outperforms the solver's default configurations, improving solution quality for ACE in 25.4% of instances and matching the default performance in 57.9%, and for Choco, achieving superior results in 38.6% of instances. It also consistently surpasses Hamming distance search within the same framework, confirming the advantage of model-based exploration over simple local search. Overall, the probe and solve algorithm offers a practical, resource-aware approach for tuning constraint solvers that yields robust improvements across diverse problem types.
Authors: Alessandro Padella, Massimiliano de Leoni, Marlon Dumas
Abstract: Predictive Process Monitoring is a branch of process mining that aims to predict the outcome of an ongoing process. Recently, it leveraged machine-and-deep learning architectures. In this paper, we extend our prior LLM-based Predictive Process Monitoring framework, which was initially focused on total time prediction via prompting. The extension consists of comprehensively evaluating its generality, semantic leverage, and reasoning mechanisms, also across multiple Key Performance Indicators. Empirical evaluations conducted on three distinct event logs and across the Key Performance Indicators of Total Time and Activity Occurrence prediction indicate that, in data-scarce settings with only 100 traces, the LLM surpasses the benchmark methods. Furthermore, the experiments also show that the LLM exploits both its embodied prior knowledge and the internal correlations among training traces. Finally, we examine the reasoning strategies employed by the model, demonstrating that the LLM does not merely replicate existing predictive methods but performs higher-order reasoning to generate the predictions.
Authors: Yohai Trabelsi, Guojun Xiong, Fentabil Getnet, St\'ephane Verguet, Milind Tambe
Abstract: Ethiopia's Ministry of Health is upgrading health posts to improve access to essential services, particularly in rural areas. Limited resources, however, require careful prioritization of which facilities to upgrade to maximize population coverage while accounting for diverse expert and stakeholder preferences. In collaboration with the Ethiopian Public Health Institute and Ministry of Health, we propose a hybrid framework that systematically integrates expert knowledge with optimization techniques. Classical optimization methods provide theoretical guarantees but require explicit, quantitative objectives, whereas stakeholder criteria are often articulated in natural language and difficult to formalize. To bridge these domains, we develop the Large language model and Extended Greedy (LEG) framework. Our framework combines a provable approximation algorithm for population coverage optimization with LLM-driven iterative refinement that incorporates human-AI alignment to ensure solutions reflect expert qualitative guidance while preserving coverage guarantees. Experiments on real-world data from three Ethiopian regions demonstrate the framework's effectiveness and its potential to inform equitable, data-driven health system planning.
Authors: Kaiwen Wang, Kaili Zheng, Rongrong Deng, Qingmin Fan, Milin Zhang, Zongrui Li, Xuesi Zhou, Bo Han, Liren Chen, Chenyi Guo, Ji Wu
Abstract: Competitive sports require sophisticated tactical analysis, yet combat disciplines like boxing remain underdeveloped in AI-driven analytics due to the complexity of action dynamics and the lack of structured tactical representations. To address this, we present BoxMind, a closed-loop AI expert system validated in elite boxing competition. By defining atomic punch events with precise temporal boundaries and spatial and technical attributes, we parse match footage into 18 hierarchical technical-tactical indicators. We then propose a graph-based predictive model that fuses these explicit technical-tactical profiles with learnable, time-variant latent embeddings to capture the dynamics of boxer matchups. Modeling match outcome as a differentiable function of technical-tactical indicators, we turn winning probability gradients into executable tactical adjustments. Experiments show that the outcome prediction model achieves state-of-the-art performance, with 69.8% accuracy on BoxerGraph test set and 87.5% on Olympic matches. Using this predictive model as a foundation, the system generates strategic recommendations that demonstrate proficiency comparable to human experts. BoxMind is validated through a closed-loop deployment during the 2024 Paris Olympics, directly contributing to the Chinese National Team's historic achievement of three gold and two silver medals. BoxMind establishes a replicable paradigm for transforming unstructured video data into strategic intelligence, bridging the gap between computer vision and decision support in competitive sports.
Authors: Thomas D. Hull, Lizhe Zhang, Patricia A. Arean, Matteo Malgaroli
Abstract: Generative artificial intelligence (GAI) chatbots built for mental health could deliver safe, personalized, and scalable mental health support. We evaluate a foundation model designed for mental health. Adults completed mental health measures while engaging with the chatbot between May 15, 2025 and September 15, 2025. Users completed an opt-in consent, demographic information, mental health symptoms, social connection, and self-identified goals. Measures were repeated every two weeks up to 6 weeks, and a final follow-up at 10 weeks. Analyses included effect sizes, and growth mixture models to identify participant groups and their characteristic engagement, severity, and demographic factors. Users demonstrated significant reductions in PHQ-9 and GAD-7 that were sustained at follow-up. Significant improvements in Hope, Behavioral Activation, Social Interaction, Loneliness, and Perceived Social Support were observed throughout and maintained at 10 week follow-up. Engagement was high and predicted outcomes. Working alliance was comparable to traditional care and predicted outcomes. Automated safety guardrails functioned as designed, with 76 sessions flagged for risk and all handled according to escalation policies. This single arm naturalistic observational study provides initial evidence that a GAI foundation model for mental health can deliver accessible, engaging, effective, and safe mental health support. These results lend support to findings from early randomized designs and offer promise for future study of mental health GAI in real world settings.
Authors: Huanxiang Lin, Qianyue Wang, Jinwu Hu, Bailin Chen, Qing Du, Mingkui Tan
Abstract: Data-driven reports communicate decision-relevant insights by tightly interleaving narrative text with charts grounded in underlying tables. However, current LLM-based systems typically generate narratives and visualizations in staged pipelines, following either a text-first-graph-second or a graph-first-text-second paradigm. These designs often lead to chart-text inconsistency and insight freezing, where the intermediate evidence space becomes fixed and the model can no longer retrieve or construct new visual evidence as the narrative evolves, resulting in shallow and predefined analysis. To address the limitations, we propose \textbf{EvidFuse}, a training-free multi-agent framework that enables writing-time text-chart interleaved generation for data-driven reports. EvidFuse decouples visualization analysis from long-form drafting via two collaborating components: a \textbf{Data-Augmented Analysis Agent}, equipped with Exploratory Data Analysis (EDA)-derived knowledge and access to raw tables, and a \textbf{Real-Time Evidence Construction Writer} that plans an outline and drafts the report while intermittently issuing fine-grained analysis requests. This design allows visual evidence to be constructed and incorporated exactly when the narrative requires it, directly constraining subsequent claims and enabling on-demand expansion of the evidence space. Experiments demonstrate that EvidFuse attains the top rank in both LLM-as-a-judge and human evaluations on chart quality, chart-text alignment, and report-level usefulness.
Authors: Hanlin Zhang, Daxin Tan, Dehua Tao, Xiao Chen, Haochen Tan, Yunhe Li, Yuchen Cao, Jianping Wang, Linqi Song
Abstract: Speech tokenizers serve as the cornerstone of discrete Speech Large Language Models (Speech LLMs). Existing tokenizers either prioritize semantic encoding, fuse semantic content with acoustic style inseparably, or achieve incomplete semantic-acoustic disentanglement. To achieve better disentanglement, we propose DSA-Tokenizer, which explicitly disentangles speech into discrete semantic and acoustic tokens via distinct optimization constraints. Specifically, semantic tokens are supervised by ASR to capture linguistic content, while acoustic tokens focus on mel-spectrograms restoration to encode style. To eliminate rigid length constraints between the two sequences, we introduce a hierarchical Flow-Matching decoder that further improve speech generation quality. Furthermore, We employ a joint reconstruction-recombination training strategy to enforce this separation. DSA-Tokenizer enables high fidelity reconstruction and flexible recombination through robust disentanglement, facilitating controllable generation in speech LLMs. Our analysis highlights disentangled tokenization as a pivotal paradigm for future speech modeling. Audio samples are avaialble at https://anonymous.4open.science/w/DSA_Tokenizer_demo/. The code and model will be made publicly available after the paper has been accepted.
URLs: https://anonymous.4open.science/w/DSA_Tokenizer_demo/.
Authors: Jakob Struye, Nabeel Nisar Bhat, Siddhartha Kumar, Mohammad Hossein Moghaddam, Jeroen Famaey
Abstract: Most Integrated Sensing and Communications (ISAC) systems require dividing airtime across their two modes. However, the specific impact of this decision on sensing performance remains unclear and underexplored. In this paper, we therefore investigate the impact on a gesture recognition system using a Millimeter-Wave (mmWave) ISAC system. With our dataset of power per beam pair gathered with two mmWave devices performing constant beam sweeps while test subjects performed distinct gestures, we train a gesture classifier using Convolutional Neural Networks. We then subsample these measurements, emulating reduced sensing airtime, showing that a sensing airtime of 25 % only reduces classification accuracy by 0.15 percentage points from full-time sensing. Alongside this high-quality sensing at low airtime, mmWave systems are known to provide extremely high data throughputs, making mmWave ISAC a prime enabler for applications such as truly wireless Extended Reality.
Authors: Anas Hajbi
Abstract: Modern neural networks rely on generic activation functions (ReLU, GELU, SiLU) that ignore the mathematical structure inherent in scientific data. We propose Neuro-Symbolic Activation Discovery, a framework that uses Genetic Programming to extract interpretable mathematical formulas from data and inject them as custom activation functions. Our key contribution is the discovery of a Geometric Transfer phenomenon: activation functions learned from particle physics data successfully generalize to ecological classification, outperforming standard activations (ReLU, GELU, SiLU) in both accuracy and parameter efficiency. On the Forest Cover dataset, our Hybrid Transfer model achieves 82.4% accuracy with only 5,825 parameters, compared to 83.4% accuracy requiring 31,801 parameters for a conventional heavy network -- a 5.5x parameter reduction with only 1% accuracy loss. We introduce a Parameter Efficiency Score ($E_{param} = AUC / \log_{10}(Params)$) and demonstrate that lightweight hybrid architectures consistently achieve 18-21% higher efficiency than over-parameterized baselines. Crucially, we establish boundary conditions: while Physics to Ecology transfer succeeds (both involve continuous Euclidean measurements), Physics to Text transfer fails (discrete word frequencies require different mathematical structures). Our work opens pathways toward domain-specific activation libraries for efficient scientific machine learning.
Authors: Am\'elie Gruel, Pierre Lewden, Adrien F. Vincent, Sylvain Sa\"ighi
Abstract: Neuromorphic vision made significant progress in recent years, thanks to the natural match between spiking neural networks and event data in terms of biological inspiration, energy savings, latency and memory use for dynamic visual data processing. However, optimising its energy requirements still remains a challenge within the community, especially for embedded applications. One solution may reside in preprocessing events to optimise data quantity thus lowering the energy cost on neuromorphic hardware, proportional to the number of synaptic operations. To this end, we extend an end-to-end neuromorphic line detection mechanism to introduce line-based event data preprocessing. Our results demonstrate on three benchmark event-based datasets that preprocessing leads to an advantageous trade-off between energy consumption and classification performance. Depending on the line-based preprocessing strategy and the complexity of the classification task, we show that one can maintain or increase the classification accuracy while significantly reducing the theoretical energy consumption. Our approach systematically leads to a significant improvement of the neuromorphic classification efficiency, thus laying the groundwork towards a more frugal neuromorphic computer vision thanks to event preprocessing.
Authors: Jun Li, Hongling Zhu, Yujie Xiao, Qinghao Zhao, Yalei Ke, Gongzheng Tang, Guangkun Nie, Deyun Zhang, Jin Li, Canqing Yu, Shenda Hong
Abstract: Background: Artificial intelligence enabled electrocardiography (AI-ECG) has demonstrated the ability to detect diverse pathologies, but most existing models focus on single disease identification, neglecting comorbidities and future risk prediction. Although ECGFounder expanded cardiac disease coverage, a holistic health profiling model remains needed. Methods: We constructed a large multicenter dataset comprising 13.3 million ECGs from 2.98 million patients. Using transfer learning, ECGFounder was fine-tuned to develop AnyECG, a foundation model for holistic health profiling. Performance was evaluated using external validation cohorts and a 10-year longitudinal cohort for current diagnosis, future risk prediction, and comorbidity identification. Results: AnyECG demonstrated systemic predictive capability across 1172 conditions, achieving an AUROC greater than 0.7 for 306 diseases. The model revealed novel disease associations, robust comorbidity patterns, and future disease risks. Representative examples included high diagnostic performance for hyperparathyroidism (AUROC 0.941), type 2 diabetes (0.803), Crohn disease (0.817), lymphoid leukemia (0.856), and chronic obstructive pulmonary disease (0.773). Conclusion: The AnyECG foundation model provides substantial evidence that AI-ECG can serve as a systemic tool for concurrent disease detection and long-term risk prediction.
Authors: Runyuan Cai, Yu Lin, Yiming Wang, Chunlin Fu, Xiaodong Zeng
Abstract: Traditional speech systems typically rely on separate, task-specific models for text-to-speech (TTS), automatic speech recognition (ASR), and voice conversion (VC), resulting in fragmented pipelines that limit scalability, efficiency, and cross-task generalization. In this paper, we present General-Purpose Audio (GPA), a unified audio foundation model that integrates multiple core speech tasks within a single large language model (LLM) architecture. GPA operates on a shared discrete audio token space and supports instruction-driven task induction, enabling a single autoregressive model to flexibly perform TTS, ASR, and VC without architectural modifications. This unified design combines a fully autoregressive formulation over discrete speech tokens, joint multi-task training across speech domains, and a scalable inference pipeline that achieves high concurrency and throughput. The resulting model family supports efficient multi-scale deployment, including a lightweight 0.3B-parameter variant optimized for edge and resource-constrained environments. Together, these design choices demonstrate that a unified autoregressive architecture can achieve competitive performance across diverse speech tasks while remaining viable for low-latency, practical deployment.
Authors: Niko Usai, Dario Montagnini, Kristian Ilianov Iliev, Raffaele Camanzo
Abstract: Understanding large software systems is a challenging task, especially when code is distributed across multiple repositories and microservices. Developers often need to reason not only about the structure of the code, but also about its domain logic and runtime behaviors, which are typically implicit and scattered. We introduce LogicLens, a reactive conversational agent that assists developers in exploring complex software systems through a semantic multi-repository graph. This graph is built in a preprocessing step by combining syntactic code analysis, via AST parsing and repository traversal, with semantic enrichment using Large Language Models (LLMs). The resulting graph captures both structural elements, such as files, classes, and functions, as well as functional abstractions like domain entities, operations, and workflows. Once the graph is constructed, LogicLens enables developers to interact with it via natural language, dynamically retrieving relevant subgraphs and answering technical or functional queries. We present the architecture of the system, discuss emergent behaviors, and evaluate its effectiveness on real-world multi-repository scenarios. We demonstrate emergent capabilities including impact analysis and symptom-based debugging that arise naturally from the semantic graph structure.
Authors: Qingyue Zhang, Chang Chu, Haohao Fu, Tianren Peng, Yanru Wu, Guanbo Huang, Yang Li, Shao-Lun Huang
Abstract: Transfer learning plays a vital role in improving model performance in data-scarce scenarios. However, naive uniform transfer from multiple source tasks may result in negative transfer, highlighting the need to properly balance the contributions of heterogeneous sources. Moreover, existing transfer learning methods typically focus on optimizing either the source weights or the amount of transferred samples, while largely neglecting the joint consideration of the other. In this work, we propose a theoretical framework, Unified Optimization of Weights and Quantities (UOWQ), which formulates multi-source transfer learning as a parameter estimation problem grounded in an asymptotic analysis of a Kullback-Leibler divergence-based generalization error measure. The proposed framework jointly determines the optimal source weights and optimal transfer quantities for each source task. Firstly, we prove that using all available source samples is always optimal once the weights are properly adjusted, and we provide a theoretical explanation for this phenomenon. Moreover, to determine the optimal transfer weights, our analysis yields closed-form solutions in the single-source setting and develops a convex optimization-based numerical procedure for the multi-source case. Building on the theoretical results, we further propose practical algorithms for both multi-source transfer learning and multi-task learning settings. Extensive experiments on real-world benchmarks, including DomainNet and Office-Home, demonstrate that UOWQ consistently outperforms strong baselines. The results validate both the theoretical predictions and the practical effectiveness of our framework.
Authors: Mengmeng Peng, Zhenyu Fang, He Sun
Abstract: Large language models (LLMs) currently suffer from parameter entanglement, where general reasoning capabilities (logic) and specific factual knowledge (facts) exist in a superposition state within shared weights. This coupling leads to the "memory wall," where computational capacity is squandered on simulating retrieval, often resulting in hallucinations. In this paper, we propose "digital metabolism," a thermodynamic hypothesis suggesting that targeted forgetting is necessary for distilling a pure neural logic core. To validate this hypothesis, we introduce the Regenerative Logic-Core Protocol (RLCP), a dual-stream training framework that renders specific factual dependencies linearly undecodable via deep-layer gradient reversal. Applying RLCP to Qwen2.5-0.5B, we observe a distinct phase transition: the model achieves near-zero retention of targeted factual associations (Accuracy < 7%) while exhibiting changes consistent with an emergent "structural crystallization" effect. Empirical analysis on GSM8K reveals that the "metabolized" model spontaneously adopts chain-of-thought (CoT) scaffolding, which we interpret as compensating for the loss of direct associative recall (shifting from $O(1)$ recall to $O(N)$ reasoning). While the causal mechanism underlying this behavioral shift requires further investigation, our findings provide a dynamic weight-level counterpart to architectural innovations like DeepSeek's Engram, paving the way for modular "Neural CPU + Symbolic RAM" architectures.
Authors: Himanshu Thakur, Anusha Kamath, Anurag Muthyala, Dhwani Sanmukhani, Smruthi Mukund, Jay Katukuri
Abstract: Recent advances in code generation models have unlocked unprecedented opportunities for automating feature engineering, yet their adoption in real-world ML teams remains constrained by critical challenges: (i) the scarcity of datasets capturing the iterative and complex coding processes of production-level feature engineering, (ii) limited integration and personalization of widely used coding agents, such as CoPilot and Devin, with a team's unique tools, codebases, workflows, and practices, and (iii) suboptimal human-AI collaboration due to poorly timed or insufficient feedback. We address these challenges with a planner-guided, constrained-topology multi-agent framework that generates code for repositories in a multi-step fashion. The LLM-powered planner leverages a team's environment, represented as a graph, to orchestrate calls to available agents, generate context-aware prompts, and use downstream failures to retroactively correct upstream artifacts. It can request human intervention at critical steps, ensuring generated code is reliable, maintainable, and aligned with team expectations. On a novel in-house dataset, our approach achieves 38% and 150% improvement in the evaluation metric over manually crafted and unplanned workflows respectively. In practice, when building features for recommendation models serving over 120 million users, our approach has delivered real-world impact by reducing feature engineering cycles from three weeks to a single day.
Authors: Simin Liu, Tong Zhao, Bernhard Paus Graesdal, Peter Werner, Jiuguang Wang, John Dolan, Changliu Liu, Tao Pang
Abstract: If we consider human manipulation, it is clear that contact-rich manipulation (CRM)-the ability to use any surface of the manipulator to make contact with objects-can be far more efficient and natural than relying solely on end-effectors (i.e., fingertips). However, state-of-the-art model-based planners for CRM are still focused on feasibility rather than optimality, limiting their ability to fully exploit CRM's advantages. We introduce a new paradigm that computes approximately optimal manipulator plans. This approach has two phases. Offline, we construct a graph of mutual reachable sets, where each set contains all object orientations reachable from a starting object orientation and grasp. Online, we plan over this graph, effectively computing and sequencing local plans for globally optimized motion. On a challenging, representative contact-rich task, our approach outperforms a leading planner, reducing task cost by 61%. It also achieves a 91% success rate across 250 queries and maintains sub-minute query times, ultimately demonstrating that globally optimized contact-rich manipulation is now practical for real-world tasks.
Authors: Hieu Bui, Nathaniel E. Chodosh, Arash Tavakoli
Abstract: As robotics become increasingly integrated into construction workflows, their ability to interpret and respond to human behavior will be essential for enabling safe and effective collaboration. Vision-Language Models (VLMs) have emerged as a promising tool for visual understanding tasks and offer the potential to recognize human behaviors without extensive domain-specific training. This capability makes them particularly appealing in the construction domain, where labeled data is scarce and monitoring worker actions and emotional states is critical for safety and productivity. In this study, we evaluate the performance of three leading VLMs, GPT-4o, Florence 2, and LLaVa-1.5, in detecting construction worker actions and emotions from static site images. Using a curated dataset of 1,000 images annotated across ten action and ten emotion categories, we assess each model's outputs through standardized inference pipelines and multiple evaluation metrics. GPT-4o consistently achieved the highest scores across both tasks, with an average F1-score of 0.756 and accuracy of 0.799 in action recognition, and an F1-score of 0.712 and accuracy of 0.773 in emotion recognition. Florence 2 performed moderately, with F1-scores of 0.497 for action and 0.414 for emotion, while LLaVa-1.5 showed the lowest overall performance, with F1-scores of 0.466 for action and 0.461 for emotion. Confusion matrix analyses revealed that all models struggled to distinguish semantically close categories, such as collaborating in teams versus communicating with supervisors. While the results indicate that general-purpose VLMs can offer a baseline capability for human behavior recognition in construction environments, further improvements, such as domain adaptation, temporal modeling, or multimodal sensing, may be needed for real-world reliability.
Authors: Chongcong Jiang, Tianxingjian Ding, Chuhan Song, Jiachen Tu, Ziyang Yan, Yihua Shao, Zhenyi Wang, Yuzhang Shang, Tianyu Han, Yu Tian
Abstract: Promptable segmentation foundation models such as SAM3 have demonstrated strong generalization capabilities through interactive and concept-based prompting. However, their direct applicability to medical image segmentation remains limited by severe domain shifts, the absence of privileged spatial prompts, and the need to reason over complex anatomical and volumetric structures. Here we present Medical SAM3, a foundation model for universal prompt-driven medical image segmentation, obtained by fully fine-tuning SAM3 on large-scale, heterogeneous 2D and 3D medical imaging datasets with paired segmentation masks and text prompts. Through a systematic analysis of vanilla SAM3, we observe that its performance degrades substantially on medical data, with its apparent competitiveness largely relying on strong geometric priors such as ground-truth-derived bounding boxes. These findings motivate full model adaptation beyond prompt engineering alone. By fine-tuning SAM3's model parameters on 33 datasets spanning 10 medical imaging modalities, Medical SAM3 acquires robust domain-specific representations while preserving prompt-driven flexibility. Extensive experiments across organs, imaging modalities, and dimensionalities demonstrate consistent and significant performance gains, particularly in challenging scenarios characterized by semantic ambiguity, complex morphology, and long-range 3D context. Our results establish Medical SAM3 as a universal, text-guided segmentation foundation model for medical imaging and highlight the importance of holistic model adaptation for achieving robust prompt-driven segmentation under severe domain shift. Code and model will be made available at https://github.com/AIM-Research-Lab/Medical-SAM3.
Authors: Pouya Afshin, David Helminiak, Tianling Niu, Julie M. Jorns, Tina Yen, Bing Yu, Dong Hye Ye
Abstract: Breast-Conserving Surgery (BCS) requires precise intraoperative margin assessment to preserve healthy tissue. Deep Ultraviolet Fluorescence Scanning Microscopy (DUV-FSM) offers rapid, high-resolution surface imaging for this purpose; however, the scarcity of annotated DUV data hinders the training of robust deep learning models. To address this, we propose an Self-Supervised Learning (SSL)-guided Latent Diffusion Model (LDM) to generate high-quality synthetic training patches. By guiding the LDM with embeddings from a fine-tuned DINO teacher, we inject rich semantic details of cellular structures into the synthetic data. We combine real and synthetic patches to fine-tune a Vision Transformer (ViT), utilizing patch prediction aggregation for WSI-level classification. Experiments using 5-fold cross-validation demonstrate that our method achieves 96.47 % accuracy and reduces the FID score to 45.72, significantly outperforming class-conditioned baselines.
Authors: Tasneem Shaffee, Sherief Reda
Abstract: Robust Multi-Task Learning (MTL) is crucial for autonomous systems operating in real-world environments, where adverse weather conditions can severely degrade model performance and reliability. In this paper, we introduce RobuMTL, a novel architecture designed to adaptively address visual degradation by dynamically selecting task-specific hierarchical Low-Rank Adaptation (LoRA) modules and a LoRA expert squad based on input perturbations in a mixture-of-experts fashion. Our framework enables adaptive specialization based on input characteristics, improving robustness across diverse real-world conditions. To validate our approach, we evaluated it on the PASCAL and NYUD-v2 datasets and compared it against single-task models, standard MTL baselines, and state-of-the-art methods. On the PASCAL benchmark, RobuMTL delivers a +2.8% average relative improvement under single perturbations and up to +44.4% under mixed weather conditions compared to the MTL baseline. On NYUD-v2, RobuMTL achieves a +9.7% average relative improvement across tasks. The code is available at GitHub.
Authors: Dustin S. Stoltz, Marshall A. Taylor, Sanuj Kumar
Abstract: Currently, there are thousands of large pretrained language models (LLMs) available to social scientists. How do we select among them? Using validity, reliability, reproducibility, and replicability as guides, we explore the significance of: (1) model openness, (2) model footprint, (3) training data, and (4) model architectures and fine-tuning. While ex-ante tests of validity (i.e., benchmarks) are often privileged in these discussions, we argue that social scientists cannot altogether avoid validating computational measures (ex-post). Replicability, in particular, is a more pressing guide for selecting language models. Being able to reliably replicate a particular finding that entails the use of a language model necessitates reliably reproducing a task. To this end, we propose starting with smaller, open models, and constructing delimited benchmarks to demonstrate the validity of the entire computational pipeline.
Authors: David Szczecina, Hudson Sun, Anthony Bertnyk, Niloofar Azad, Kyle Gao, Lincoln Linlin Xu
Abstract: Tree canopy detection from aerial imagery is an important task for environmental monitoring, urban planning, and ecosystem analysis. Simulating real-life data annotation scarcity, the Solafune Tree Canopy Detection competition provides a small and imbalanced dataset of only 150 annotated images, posing significant challenges for training deep models without severe overfitting. In this work, we evaluate five representative architectures, YOLOv11, Mask R-CNN, DeepLabv3, Swin-UNet, and DINOv2, to assess their suitability for canopy segmentation under extreme data scarcity. Our experiments show that pretrained convolution-based models, particularly YOLOv11 and Mask R-CNN, generalize significantly better than pretrained transformer-based models. DeeplabV3, Swin-UNet and DINOv2 underperform likely due to differences between semantic and instance segmentation tasks, the high data requirements of Vision Transformers, and the lack of strong inductive biases. These findings confirm that transformer-based architectures struggle in low-data regimes without substantial pretraining or augmentation and that differences between semantic and instance segmentation further affect model performance. We provide a detailed analysis of training strategies, augmentation policies, and model behavior under the small-data constraint and demonstrate that lightweight CNN-based methods remain the most reliable for canopy detection on limited imagery.
Authors: K Lokesh, Abhirama Subramanyam Penamakuri, Uday Agarwal, Apoorva Challa, Shreya K Gowda, Somesh Gupta, Anand Mishra
Abstract: Traditionally, AI research in medical diagnosis has largely centered on image analysis. While this has led to notable advancements, the absence of patient-reported symptoms continues to hinder diagnostic accuracy. To address this, we propose a Pre-Consultation Dialogue Framework (PCDF) that mimics real-world diagnostic procedures, where doctors iteratively query patients before reaching a conclusion. Specifically, we simulate diagnostic dialogues between two vision-language models (VLMs): a DocVLM, which generates follow-up questions based on the image and dialogue history, and a PatientVLM, which responds using a symptom profile derived from the ground-truth diagnosis. We additionally conducted a small-scale clinical validation of the synthetic symptoms generated by our framework, with licensed clinicians confirming their clinical relevance, symptom coverage, and overall realism. These findings indicate that the resulting DocVLM-PatientVLM interactions form coherent, multi-turn consultations paired with images and diagnoses, which we then use to fine-tune the DocVLM. This dialogue-based supervision leads to substantial gains over image-only training, highlighting the value of realistic symptom elicitation for diagnosis.
Authors: Shijie Jiang, Zefan Zhang, Kehua Zhu, Tian Bai, Ruihong Zhao
Abstract: The simulation of realistic clinical interactions plays a pivotal role in advancing clinical Large Language Models (LLMs) and supporting medical diagnostic education. Existing approaches and benchmarks rely on generic or LLM-generated dialogue data, which limits the authenticity and diversity of doctor-patient interactions. In this work, we propose the first Chinese patient simulation dataset (Ch-PatientSim), constructed from realistic clinical interaction scenarios to comprehensively evaluate the performance of models in emulating patient behavior. Patients are simulated based on a five-dimensional persona structure. To address issues of the persona class imbalance, a portion of the dataset is augmented using few-shot generation, followed by manual verification. We evaluate various state-of-the-art LLMs and find that most produce overly formal responses that lack individual personality. To address this limitation, we propose a training-free Multi-Stage Patient Role-Playing (MSPRP) framework, which decomposes interactions into three stages to ensure both personalization and realism in model responses. Experimental results demonstrate that our approach significantly improves model performance across multiple dimensions of patient simulation.
Authors: Kaiyu Zhou, Yongsen Zheng, Yicheng He, Meng Xue, Xueluan Gong, Yuji Wang, Kwok-Yan Lam
Abstract: The agent-tool communication loop is a critical attack surface in modern Large Language Model (LLM) agents. Existing Denial-of-Service (DoS) attacks, primarily triggered via user prompts or injected retrieval-augmented generation (RAG) context, are ineffective for this new paradigm. They are fundamentally single-turn and often lack a task-oriented approach, making them conspicuous in goal-oriented workflows and unable to exploit the compounding costs of multi-turn agent-tool interactions. We introduce a stealthy, multi-turn economic DoS attack that operates at the tool layer under the guise of a correctly completed task. Our method adjusts text-visible fields and a template-governed return policy in a benign, Model Context Protocol (MCP)-compatible tool server, optimizing these edits with a Monte Carlo Tree Search (MCTS) optimizer. These adjustments leave function signatures unchanged and preserve the final payload, steering the agent into prolonged, verbose tool-calling sequences using text-only notices. This compounds costs across turns, escaping single-turn caps while keeping the final answer correct to evade validation. Across six LLMs on the ToolBench and BFCL benchmarks, our attack expands tasks into trajectories exceeding 60,000 tokens, inflates costs by up to 658x, and raises energy by 100-560x. It drives GPU KV cache occupancy from <1% to 35-74% and cuts co-running throughput by approximately 50%. Because the server remains protocol-compatible and task outcomes are correct, conventional checks fail. These results elevate the agent-tool interface to a first-class security frontier, demanding a paradigm shift from validating final answers to monitoring the economic and computational cost of the entire agentic process.
Authors: Hyeseon An, Shinwoo Park, Hyundong Jin, Yo-Sub Han
Abstract: Steering LLMs is essential for specialized applications such as style-sensitive text rewriting, user-adaptive communication, and toxicity mitigation. Current steering methods, such as prompting-based and activation-based approaches, are widely used to guide model behavior. However, activation-based techniques require deep access to internal layers, while prompting-based steering often fails to provide consistent or fine-grained control. In order to address these limitations, we propose a training-free inference-time logit intervention for controllable generation. Our approach utilizes a statistical token score table derived from z-normalized log-odds of labeled corpora to shift the decoding distribution. Empirical evaluations across three diverse datasets focusing on writing complexity, formality, and toxicity demonstrate that our method effectively steers output characteristics, confirming its broad applicability and task-agnostic nature. Our results show that statistically grounded logit steering can achieve large, consistent, and multi-task control gains: up to +47%p accuracy and 50x f1 improvement.
Authors: Zhongxiang Sun, Yi Zhan, Chenglei Shen, Weijie Yu, Xiao Zhang, Ming He, Jun Xu
Abstract: Personalized large language models (LLMs) adapt model behavior to individual users to enhance user satisfaction, yet personalization can inadvertently distort factual reasoning. We show that when personalized LLMs face factual queries, there exists a phenomenon where the model generates answers aligned with a user's prior history rather than the objective truth, resulting in personalization-induced hallucinations that degrade factual reliability and may propagate incorrect beliefs, due to representational entanglement between personalization and factual representations. To address this issue, we propose Factuality-Preserving Personalized Steering (FPPS), a lightweight inference-time approach that mitigates personalization-induced factual distortions while preserving personalized behavior. We further introduce PFQABench, the first benchmark designed to jointly evaluate factual and personalized question answering under personalization. Experiments across multiple LLM backbones and personalization methods show that FPPS substantially improves factual accuracy while maintaining personalized performance.
Authors: Fenglin Zhang, Jie Wang
Abstract: In this paper, we introduce a framework for contextual distributionally robust optimization (DRO) that considers the causal and continuous structure of the underlying distribution by developing interpretable and tractable decision rules that prescribe decisions using covariates. We first introduce the causal Sinkhorn discrepancy (CSD), an entropy-regularized causal Wasserstein distance that encourages continuous transport plans while preserving the causal consistency. We then formulate a contextual DRO model with a CSD-based ambiguity set, termed Causal Sinkhorn DRO (Causal-SDRO), and derive its strong dual reformulation where the worst-case distribution is characterized as a mixture of Gibbs distributions. To solve the corresponding infinite-dimensional policy optimization, we propose the Soft Regression Forest (SRF) decision rule, which approximates optimal policies within arbitrary measurable function spaces. The SRF preserves the interpretability of classical decision trees while being fully parametric, differentiable, and Lipschitz smooth, enabling intrinsic interpretation from both global and local perspectives. To solve the Causal-SDRO with parametric decision rules, we develop an efficient stochastic compositional gradient algorithm that converges to an $\varepsilon$-stationary point at a rate of $O(\varepsilon^{-4})$, matching the convergence rate of standard stochastic gradient descent. Finally, we validate our method through numerical experiments on synthetic and real-world datasets, demonstrating its superior performance and interpretability.
Authors: Xinwei Wu, Heng Liu, Xiaohu Zhao, Yuqi Ren, Linlong Xu, Longyue Wang, Deyi Xiong, Weihua Luo, Kaifu Zhang
Abstract: Large Language Models (LLMs) frequently exhibit strong translation abilities, even without task-specific fine-tuning. However, the internal mechanisms governing this innate capability remain largely opaque. To demystify this process, we leverage Sparse Autoencoders (SAEs) and introduce a novel framework for identifying task-specific features. Our method first recalls features that are frequently co-activated on translation inputs and then filters them for functional coherence using a PCA-based consistency metric. This framework successfully isolates a small set of **translation initiation** features. Causal interventions demonstrate that amplifying these features steers the model towards correct translation, while ablating them induces hallucinations and off-task outputs, confirming they represent a core component of the model's innate translation competency. Moving from analysis to application, we leverage this mechanistic insight to propose a new data selection strategy for efficient fine-tuning. Specifically, we prioritize training on **mechanistically hard** samples-those that fail to naturally activate the translation initiation features. Experiments show this approach significantly improves data efficiency and suppresses hallucinations. Furthermore, we find these mechanisms are transferable to larger models of the same family. Our work not only decodes a core component of the translation mechanism in LLMs but also provides a blueprint for using internal model mechanism to create more robust and efficient models. The codes are available at https://github.com/flamewei123/AAAI26-translation-Initiation-Features.
URLs: https://github.com/flamewei123/AAAI26-translation-Initiation-Features.
Authors: Kecheng Cai, Chenyang Xu, Chao Peng
Abstract: Interpretable graph learning has recently emerged as a popular research topic in machine learning. The goal is to identify the important nodes and edges of an input graph that are crucial for performing a specific graph reasoning task. A number of studies have been conducted in this area, and various benchmark datasets have been proposed to facilitate evaluation. Among them, one of the most challenging is the Spurious-Motif benchmark, introduced at ICLR 2022. The datasets in this synthetic benchmark are deliberately designed to include spurious correlations, making it particularly difficult for models to distinguish truly relevant structures from misleading patterns. As a result, existing methods exhibit significantly worse performance on this benchmark compared to others. In this paper, we focus on improving interpretability on the challenging Spurious-Motif datasets. We demonstrate that the self-reflection technique, commonly used in large language models to tackle complex tasks, can also be effectively adapted to enhance interpretability in datasets with strong spurious correlations. Specifically, we propose a self-reflection framework that can be integrated with existing interpretable graph learning methods. When such a method produces importance scores for each node and edge, our framework feeds these predictions back into the original method to perform a second round of evaluation. This iterative process mirrors how large language models employ self-reflective prompting to reassess their previous outputs. We further analyze the reasons behind this improvement from the perspective of graph representation learning, which motivates us to propose a fine-tuning training method based on this feedback mechanism.
Authors: Xianliang Huang, Jiajie Gou, Shuhang Chen, Zhizhou Zhong, Jihong Guan, Shuigeng Zhou
Abstract: This paper presents the first unified distractor removal method, named IDDR-NGP, which directly operates on Instant-NPG. The method is able to remove a wide range of distractors in 3D scenes, such as snowflakes, confetti, defoliation and petals, whereas existing methods usually focus on a specific type of distractors. By incorporating implicit 3D representations with 2D detectors, we demonstrate that it is possible to efficiently restore 3D scenes from multiple corrupted images. We design the learned perceptual image patch similarity~( LPIPS) loss and the multi-view compensation loss (MVCL) to jointly optimize the rendering results of IDDR-NGP, which could aggregate information from multi-view corrupted images. All of them can be trained in an end-to-end manner to synthesize high-quality 3D scenes. To support the research on distractors removal in implicit 3D representations, we build a new benchmark dataset that consists of both synthetic and real-world distractors. To validate the effectiveness and robustness of IDDR-NGP, we provide a wide range of distractors with corresponding annotated labels added to both realistic and synthetic scenes. Extensive experimental results demonstrate the effectiveness and robustness of IDDR-NGP in removing multiple types of distractors. In addition, our approach achieves results comparable with the existing SOTA desnow methods and is capable of accurately removing both realistic and synthetic distractors.
Authors: Long Ma, Zihao Xue, Yan Wang, Zhiyuan Yan, Jin Xu, Xiaorui Jiang, Haiyang Yu, Yong Liao, Zhen Bi
Abstract: Recent advances in generative modeling can create remarkably realistic synthetic videos, making it increasingly difficult for humans to distinguish them from real ones and necessitating reliable detection methods. However, two key limitations hinder the development of this field. \textbf{From the dataset perspective}, existing datasets are often limited in scale and constructed using outdated or narrowly scoped generative models, making it difficult to capture the diversity and rapid evolution of modern generative techniques. Moreover, the dataset construction process frequently prioritizes quantity over quality, neglecting essential aspects such as semantic diversity, scenario coverage, and technological representativeness. \textbf{From the benchmark perspective}, current benchmarks largely remain at the stage of dataset creation, leaving many fundamental issues and in-depth analysis yet to be systematically explored. Addressing this gap, we propose AIGVDBench, a benchmark designed to be comprehensive and representative, covering \textbf{31} state-of-the-art generation models and over \textbf{440,000} videos. By executing more than \textbf{1,500} evaluations on \textbf{33} existing detectors belonging to four distinct categories. This work presents \textbf{8 in-depth analyses} from multiple perspectives and identifies \textbf{4 novel findings} that offer valuable insights for future research. We hope this work provides a solid foundation for advancing the field of AI-generated video detection. Our benchmark is open-sourced at https://github.com/LongMa-2025/AIGVDBench.
Authors: Chi Zhang, Mengqi Zhang, Xiaotian Ye, Runxi Cheng, Zisheng Zhou, Ying Zhou, Pengjie Ren, Zhumin Chen
Abstract: Sequential knowledge editing in large language models often causes catastrophic collapse of the model's general abilities, especially for parameter-modifying methods. Existing approaches mitigate this issue through heuristic constraints on parameter updates, yet the mechanisms underlying such degradation remain insufficiently understood. In this work, we present a spectral analysis of sequential knowledge editing and show that a model's general abilities are closely associated with dominant singular directions of pretrained weight matrices. These directions are highly sensitive to perturbations and are progressively disrupted by repeated edits, closely tracking the collapse in both editing efficacy and general performance. Building on this insight, we propose REVIVE, a plug-and-play framework that stabilizes sequential editing by explicitly preserving the dominant singular subspace. REVIVE represents parameter updates in the spectral basis of the original weights and filters components that would interfere with the protected region. Extensive experiments across multiple models and benchmarks show that REVIVE consistently improves editing efficacy while substantially preserving general abilities under long-horizon sequential editing, including extreme settings with up to 20,000 edits.
Authors: Stephen Pilli, Vivek Nallur
Abstract: We examine whether large language models (LLMs) can predict biased decision-making in conversational settings, and whether their predictions capture not only human cognitive biases but also how those effects change under cognitive load. In a pre-registered study (N = 1,648), participants completed six classic decision-making tasks via a chatbot with dialogues of varying complexity. Participants exhibited two well-documented cognitive biases: the Framing Effect and the Status Quo Bias. Increased dialogue complexity resulted in participants reporting higher mental demand. This increase in cognitive load selectively, but significantly, increased the effect of the biases, demonstrating the load-bias interaction. We then evaluated whether LLMs (GPT-4, GPT-5, and open-source models) could predict individual decisions given demographic information and prior dialogue. While results were mixed across choice problems, LLM predictions that incorporated dialogue context were significantly more accurate in several key scenarios. Importantly, their predictions reproduced the same bias patterns and load-bias interactions observed in humans. Across all models tested, the GPT-4 family consistently aligned with human behavior, outperforming GPT-5 and open-source models in both predictive accuracy and fidelity to human-like bias patterns. These findings advance our understanding of LLMs as tools for simulating human decision-making and inform the design of conversational agents that adapt to user biases.
Authors: Haishan Zeng, Peng Li
Abstract: In embodied artificial intelligence, enabling heterogeneous robot teams to execute long-horizon tasks from high-level instructions remains a critical challenge. While large language models (LLMs) show promise in instruction parsing and preliminary planning, they exhibit limitations in long-term reasoning and dynamic multi-robot coordination. We propose Hierarchical Autonomous Intelligent Multi-Robot Planning(H-AIM), a novel embodied multi-robot task planning framework that addresses these issues through a three-stage cascaded architecture: 1) It leverages an LLM to parse instructions and generate Planning Domain Definition Language (PDDL) problem descriptions, thereby transforming commands into formal planning problems; 2) It combines the semantic reasoning of LLMs with the search capabilities of a classical planner to produce optimized action sequences; 3) It compiles the resulting plan into behavior trees for reactive control. The framework supports dynamically sized heterogeneous robot teams via a shared blackboard mechanism for communication and state synchronization. To validate our approach, we introduce the MACE-THOR benchmark dataset, comprising 42 complex tasks across 8 distinct household layouts. Experimental results demonstrate that H-AIM achieves a remarkable performance improvement, elevating the task success rate from 12% to 55% and boosting the goal condition recall from 32% to 72% against the strongest baseline, LaMMA-P.
Authors: Rachmadita Andreswari, Stephan A. Fahrenkrog-Petersen, Jan Mendling
Abstract: Fairness in automated decision-making has become a critical concern, particularly in high-pressure healthcare scenarios such as emergency triage, where fast and equitable decisions are essential. Process mining is increasingly investigating fairness. There is a growing area focusing on fairness-aware algorithms. So far, we know less how these concepts perform on empirical healthcare data or how they cover aspects of justice theory. This study addresses this research problem and proposes a process mining approach to assess fairness in triage by linking real-life event logs with conceptual dimensions of justice. Using the MIMICEL event log (as derived from MIMIC-IV ED), we analyze time, re-do, deviation and decision as process outcomes, and evaluate the influence of age, gender, race, language and insurance using the Kruskal-Wallis, Chi-square and effect size measurements. These outcomes are mapped to justice dimensions to support the development of a conceptual framework. The results demonstrate which aspects of potential unfairness in high-acuity and sub-acute surface. In this way, this study contributes empirical insights that support further research in responsible, fairness-aware process mining in healthcare.
Authors: Rongkun Cui, Nana Zhang, Kun Zhu, Qi Zhang
Abstract: Online financial services constitute an essential component of contemporary web ecosystems, yet their openness introduces substantial exposure to fraud that harms vulnerable users and weakens trust in digital finance. Such threats have become a significant web harm that erodes societal fairness and affects the well being of online communities. However, existing detection methods based on graph neural networks (GNNs) struggle with two persistent challenges: (1) fraud camouflage, where malicious transactions mimic benign behaviors to evade detection, and (2) long-tailed data distributions, which obscure rare but critical fraudulent cases. To fill these gaps, we propose HIMVH, a Hippocampus-Inspired Multi-View Hypergraph learning model for web finance fraud detection. Specifically, drawing inspiration from the scene conflict monitoring role of the hippocampus, we design a cross-view inconsistency perception module that captures subtle discrepancies and behavioral heterogeneity across multiple transaction views. This module enables the model to identify subtle cross-view conflicts for detecting online camouflaged fraudulent behaviors. Furthermore, inspired by the match-mismatch novelty detection mechanism of the CA1 region, we introduce a novelty-aware hypergraph learning module that measures feature deviations from neighborhood expectations and adaptively reweights messages, thereby enhancing sensitivity to online rare fraud patterns in the long-tailed settings. Extensive experiments on six web-based financial fraud datasets demonstrate that HIMVH achieves 6.42\% improvement in AUC, 9.74\% in F1 and 39.14\% in AP on average over 15 SOTA models.
Authors: Jiaqi Liang, Yue Chen, Qize Yu, Yan Shen, Haipeng Zhang, Hao Dong, Ruihai Wu
Abstract: Furniture assembly is a crucial yet challenging task for robots, requiring precise dual-arm coordination where one arm manipulates parts while the other provides collaborative support and stabilization. To accomplish this task more effectively, robots need to actively adapt support strategies throughout the long-horizon assembly process, while also generalizing across diverse part geometries. We propose A3D, a framework which learns adaptive affordances to identify optimal support and stabilization locations on furniture parts. The method employs dense point-level geometric representations to model part interaction patterns, enabling generalization across varied geometries. To handle evolving assembly states, we introduce an adaptive module that uses interaction feedback to dynamically adjust support strategies during assembly based on previous interactions. We establish a simulation environment featuring 50 diverse parts across 8 furniture types, designed for dual-arm collaboration evaluation. Experiments demonstrate that our framework generalizes effectively to diverse part geometries and furniture categories in both simulation and real-world settings.
Authors: Jie Yang, Honglin Guo, Li Ji, Jiazheng Zhou, Rui Zheng, Zhikai Lei, Shuo Zhang, Zhiheng Xi, Shichun Liu, Yuxin Wang, Bo Wang, Yining Zheng, Tao Gui, Xipeng Qiu
Abstract: The evolution of Large Language Models (LLMs) into autonomous agents has expanded the scope of AI coding from localized code generation to complex, repository-level, and execution-driven problem solving. However, current benchmarks predominantly evaluate code logic in static contexts, neglecting the dynamic, full-process requirements of real-world engineering, particularly in backend development which demands rigorous environment configuration and service deployment. To address this gap, we introduce ABC-Bench, a benchmark explicitly designed to evaluate agentic backend coding within a realistic, executable workflow. Using a scalable automated pipeline, we curated 224 practical tasks spanning 8 languages and 19 frameworks from open-source repositories. Distinct from previous evaluations, ABC-Bench require the agents to manage the entire development lifecycle from repository exploration to instantiating containerized services and pass the external end-to-end API tests. Our extensive evaluation reveals that even state-of-the-art models struggle to deliver reliable performance on these holistic tasks, highlighting a substantial disparity between current model capabilities and the demands of practical backend engineering. Our code is available at https://github.com/OpenMOSS/ABC-Bench.
Authors: Jiaohong Yao, Linfeng Liang, Yao Deng, Xi Zheng, Richard Han, Yuankai Qi
Abstract: Marker-based landing is widely used in drone delivery and return-to-base systems for its simplicity and reliability. However, most approaches assume idealized landing site visibility and sensor performance, limiting robustness in complex urban settings. We present a simulation-based evaluation suite on the AirSim platform with systematically varied urban layouts, lighting, and weather to replicate realistic operational diversity. Using onboard camera sensors (RGB for marker detection and depth for obstacle avoidance), we benchmark two heuristic coverage patterns and a reinforcement learning-based agent, analyzing how exploration strategy and scene complexity affect success rate, path efficiency, and robustness. Results underscore the need to evaluate marker-based autonomous landing under diverse, sensor-relevant conditions to guide the development of reliable aerial navigation systems.
Authors: Davor Lauc
Abstract: We present a convolutional neural network approach for classifying proper names by language and entity type. Our model, Onomas-CNN X, combines parallel convolution branches with depthwise-separable operations and hierarchical classification to process names efficiently on CPU hardware. We evaluate the architecture on a large multilingual dataset covering 104 languages and four entity types (person, organization, location, other). Onomas-CNN X achieves 92.1% accuracy while processing 2,813 names per second on a single CPU core - 46 times faster than fine-tuned XLM-RoBERTa with comparable accuracy. The model reduces energy consumption by a factor of 46 compared to transformer baselines. Our experiments demonstrate that specialized CNN architectures remain competitive with large pre-trained models for focused NLP tasks when sufficient training data exists.
Authors: Shaofeng Yin, Jiaxin Ge, Zora Zhiruo Wang, Xiuyu Li, Michael J. Black, Trevor Darrell, Angjoo Kanazawa, Haiwen Feng
Abstract: Vision-as-inverse-graphics, the concept of reconstructing an image as an editable graphics program is a long-standing goal of computer vision. Yet even strong VLMs aren't able to achieve this in one-shot as they lack fine-grained spatial and physical grounding capability. Our key insight is that closing this gap requires interleaved multimodal reasoning through iterative execution and verification. Stemming from this, we present VIGA (Vision-as-Inverse-Graphic Agent) that starts from an empty world and reconstructs or edits scenes through a closed-loop write-run-render-compare-revise procedure. To support long-horizon reasoning, VIGA combines (i) a skill library that alternates generator and verifier roles and (ii) an evolving context memory that contains plans, code diffs, and render history. VIGA is task-agnostic as it doesn't require auxiliary modules, covering a wide range of tasks such as 3D reconstruction, multi-step scene editing, 4D physical interaction, and 2D document editing, etc. Empirically, we found VIGA substantially improves one-shot baselines on BlenderGym (35.32%) and SlideBench (117.17%). Moreover, VIGA is also model-agnostic as it doesn't require finetuning, enabling a unified protocol to evaluate heterogeneous foundation VLMs. To better support this protocol, we introduce BlenderBench, a challenging benchmark that stress-tests interleaved multimodal reasoning with graphics engine, where VIGA improves by 124.70%.
Authors: Xiaoyu Liang, Yuchen Peng, Jiale Luo, Wenhao Wang, Haoji Hu, Xincheng Zhou
Abstract: Large Language Models (LLMs) adapted via contrastive learning excel in general representation learning but struggle in vertical domains like chemistry and law, primarily due to a lack of domain-specific knowledge. This work identifies a core bottleneck: the prevailing ``LLM+CL'' paradigm focuses on semantic alignment but cannot perform knowledge acquisition, leading to failures on specialized terminology. To bridge this gap, we propose Learn Before Represent (LBR), a novel two-stage framework. LBR first injects domain knowledge via an Information Bottleneck-Constrained Generative Learning stage, preserving the LLM's causal attention to maximize knowledge acquisition while compressing semantics. It then performs Generative-Refined Contrastive Learning on the compressed representations for alignment. This approach maintains architectural consistency and resolves the objective conflict between generative and contrastive learning. Extensive experiments on medical, chemistry, and code retrieval tasks show that LBR significantly outperforms strong baselines. Our work establishes a new paradigm for building accurate and robust representations in vertical domains.
Authors: Van Thuy Hoang, O-Joun Lee
Abstract: Molecular property prediction is becoming one of the major applications of graph learning in Web-based services, e.g., online protein structure prediction and drug discovery. A key challenge arises in few-shot scenarios, where only a few labeled molecules are available for predicting unseen properties. Recently, several studies have used in-context learning to capture relationships among molecules and properties, but they face two limitations in: (1) exploiting prior knowledge of functional groups that are causally linked to properties and (2) identifying key substructures directly correlated with properties. We propose CaMol, a context-aware graph causality inference framework, to address these challenges by using a causal inference perspective, assuming that each molecule consists of a latent causal structure that determines a specific property. First, we introduce a context graph that encodes chemical knowledge by linking functional groups, molecules, and properties to guide the discovery of causal substructures. Second, we propose a learnable atom masking strategy to disentangle causal substructures from confounding ones. Third, we introduce a distribution intervener that applies backdoor adjustment by combining causal substructures with chemically grounded confounders, disentangling causal effects from real-world chemical variations. Experiments on diverse molecular datasets showed that CaMol achieved superior accuracy and sample efficiency in few-shot tasks, showing its generalizability to unseen properties. Also, the discovered causal substructures were strongly aligned with chemical knowledge about functional groups, supporting the model interpretability.
Authors: Minho Lee, Hyeonseok Kim, Jin Tak Kim, Sangshin Park, Jeong Hyun Lee, Jungsan Cho, Jemin Hwangbo
Abstract: The simulation-to-reality (sim-to-real) transfer of large-scale hydraulic robots presents a significant challenge in robotics because of the inherent slow control response and complex fluid dynamics. The complex dynamics result from the multiple interconnected cylinder structure and the difference in fluid rates of the cylinders. These characteristics complicate detailed simulation for all joints, making it unsuitable for reinforcement learning (RL) applications. In this work, we propose an analytical actuator model driven by hydraulic dynamics to represent the complicated actuators. The model predicts joint torques for all 12 actuators in under 1 microsecond, allowing rapid processing in RL environments. We compare our model with neural network-based actuator models and demonstrate the advantages of our model in data-limited scenarios. The locomotion policy trained in RL with our model is deployed on a hydraulic quadruped robot, which is over 300 kg. This work is the first demonstration of a successful transfer of stable and robust command-tracking locomotion with RL on a heavy hydraulic quadruped robot, demonstrating advanced sim-to-real transferability.
Authors: Yuejie Li, Ke Yang, Tao Wang, Bolin Chen, Bowen Li, Chengjun Mao
Abstract: Graph-based Retrieval-Augmented Generation (GraphRAG) frameworks face a trade-off between the comprehensiveness of global search and the efficiency of local search. Existing methods are often challenged by navigating large-scale hierarchical graphs, optimizing retrieval paths, and balancing exploration-exploitation dynamics, frequently lacking robust multi-stage re-ranking. To overcome these deficits, we propose Deep GraphRAG, a framework designed for a balanced approach to hierarchical retrieval and adaptive integration. It introduces a hierarchical global-to-local retrieval strategy that integrates macroscopic inter-community and microscopic intra-community contextual relations. This strategy employs a three-stage process: (1) inter-community filtering, which prunes the search space using local context; (2) community-level refinement, which prioritizes relevant subgraphs via entity-interaction analysis; and (3) entity-level fine-grained search within target communities. A beam search-optimized dynamic re-ranking module guides this process, continuously filtering candidates to balance efficiency and global comprehensiveness. Deep GraphRAG also features a Knowledge Integration Module leveraging a compact LLM, trained with Dynamic Weighting Reward GRPO (DW-GRPO). This novel reinforcement learning approach dynamically adjusts reward weights to balance three key objectives: relevance, faithfulness, and conciseness. This training enables compact models (1.5B) to approach the performance of large models (70B) in the integration task. Evaluations on Natural Questions and HotpotQA demonstrate that Deep GraphRAG significantly outperforms baseline graph retrieval methods in both accuracy and efficiency.
Authors: Ji Dai, Quan Fang, Jun Hu, Desheng Cai, Yang Yang, Can Zhao
Abstract: Multimedia recommendation systems leverage user-item interactions and multimodal information to capture user preferences, enabling more accurate and personalized recommendations. Despite notable advancements, existing approaches still face two critical limitations: first, shallow modality fusion often relies on simple concatenation, failing to exploit rich synergic intra- and inter-modal relationships; second, asymmetric feature treatment-where users are only characterized by interaction IDs while items benefit from rich multimodal content-hinders the learning of a shared semantic space. To address these issues, we propose a Cross-modal Recursive Attention Network with dual graph Embedding (CRANE). To tackle shallow fusion, we design a core Recursive Cross-Modal Attention (RCA) mechanism that iteratively refines modality features based on cross-correlations in a joint latent space, effectively capturing high-order intra- and inter-modal dependencies. For symmetric multimodal learning, we explicitly construct users' multimodal profiles by aggregating features of their interacted items. Furthermore, CRANE integrates a symmetric dual-graph framework-comprising a heterogeneous user-item interaction graph and a homogeneous item-item semantic graph-unified by a self-supervised contrastive learning objective to fuse behavioral and semantic signals. Despite these complex modeling capabilities, CRANE maintains high computational efficiency. Theoretical and empirical analyses confirm its scalability and high practical efficiency, achieving faster convergence on small datasets and superior performance ceilings on large-scale ones. Comprehensive experiments on four public real-world datasets validate an average 5% improvement in key metrics over state-of-the-art baselines.
Authors: Claudia Plant, Lena G. M. Bauer, Christian B\"ohm
Abstract: How to find a natural grouping of a large real data set? Clustering requires a balance between abstraction and representation. To identify clusters, we need to abstract from superfluous details of individual objects. But we also need a rich representation that emphasizes the key features shared by groups of objects that distinguish them from other groups of objects. Each clustering algorithm implements a different trade-off between abstraction and representation. Classical K-means implements a high level of abstraction - details are simply averaged out - combined with a very simple representation - all clusters are Gaussians in the original data space. We will see how approaches to subspace and deep clustering support high-dimensional and complex data by allowing richer representations. However, with increasing representational expressiveness comes the need to explicitly enforce abstraction in the objective function to ensure that the resulting method performs clustering and not just representation learning. We will see how current deep clustering methods define and enforce abstraction through centroid-based and density-based clustering losses. Balancing the conflicting goals of abstraction and representation is challenging. Ideas from subspace clustering help by learning one latent space for the information that is relevant to clustering and another latent space to capture all other information in the data. The tutorial ends with an outlook on future research in clustering. Future methods will more adaptively balance abstraction and representation to improve performance, energy efficiency and interpretability. By automatically finding the sweet spot between abstraction and representation, the human brain is very good at clustering and other related tasks such as single-shot learning. So, there is still much room for improvement.
Authors: Luisa Carpinelli, Filippo Natoli, Marco Taboga
Abstract: Artificial intelligence (AI) has moved to the center of policy, market, and academic debates, but its macroeconomic footprint is still only partly understood. This paper provides an overview on how the current AI wave is captured in US national accounts, combining a simple macro-accounting framework with a stylized description of the AI production process. We highlight the crucial role played by data centers, which constitute the backbone of the AI ecosystem and have attracted formidable investment in 2025, as they are indispensable for meeting the rapidly increasing worldwide demand for AI services. We document that the boom in IT and AI-related capital expenditure in the first three quarters of the year has given an outsized boost to aggregate demand, while its contribution to GDP growth is smaller once the high import content of AI hardware is netted out. Furthermore, simple calculations suggest that, at current utilization rates and pricing, the production of services originating in new AI data centers could contribute to GDP over the turn of the next quarters on a scale comparable to that of investment spending to date. Short reinvestment cycles and uncertainty about future AI demand, while not currently acting as a macroeconomic drag, can nevertheless fuel macroeconomic risks over the medium term.
Authors: Aiman Al Masoud, Marco Arazzi, Antonino Nocera
Abstract: Retrieval-Augmented Generation (RAG) has attracted significant attention due to its ability to combine the generative capabilities of Large Language Models (LLMs) with knowledge obtained through efficient retrieval mechanisms over large-scale data collections. Currently, the majority of existing approaches overlook the risks associated with exposing sensitive or access-controlled information directly to the generation model. Only a few approaches propose techniques to instruct the generative model to refrain from disclosing sensitive information; however, recent studies have also demonstrated that LLMs remain vulnerable to prompt injection attacks that can override intended behavioral constraints. For these reasons, we propose a novel approach to Selective Disclosure in Retrieval-Augmented Generation, called SD-RAG, which decouples the enforcement of security and privacy constraints from the generation process itself. Rather than relying on prompt-level safeguards, SD-RAG applies sanitization and disclosure controls during the retrieval phase, prior to augmenting the language model's input. Moreover, we introduce a semantic mechanism to allow the ingestion of human-readable dynamic security and privacy constraints together with an optimized graph-based data model that supports fine-grained, policy-aware retrieval. Our experimental evaluation demonstrates the superiority of SD-RAG over baseline existing approaches, achieving up to a $58\%$ improvement in the privacy score, while also showing a strong resilience to prompt injection attacks targeting the generative model.
Authors: Haiyang Xiao, Weiqing Li, Jinyue Guo, Guochao Jiang, Guohua Liu, Yuewei Zhang
Abstract: Although post-training quantization (PTQ) provides an efficient numerical compression scheme for deploying large language models (LLMs) on resource-constrained devices, the representativeness and universality of calibration data remain a core bottleneck in determining the accuracy of quantization parameters. Traditional PTQ methods typically rely on limited samples, making it difficult to capture the activation distribution during the inference phase, leading to biases in quantization parameters. To address this, we propose \textbf{FAQ} (Family-Aware Quantization), a calibration data regeneration framework that leverages prior knowledge from LLMs of the same family to generate high-fidelity calibration samples. Specifically, FAQ first inputs the original calibration samples into a larger LLM from the same family as the target model, regenerating a series of high-fidelity calibration data using a highly consistent knowledge system. Subsequently, this data, carrying Chain-of-Thought reasoning and conforming to the expected activation distribution, undergoes group competition under expert guidance to select the best samples, which are then re-normalized to enhance the effectiveness of standard PTQ. Experiments on multiple model series, including Qwen3-8B, show that FAQ reduces accuracy loss by up to 28.5\% compared to the baseline with original calibration data, demonstrating its powerful potential and contribution.
Authors: Emanuele Ratti
Abstract: The past few years have witnessed an increasing use of machine learning (ML) systems in science. Paul Humphreys has argued that, because of specific characteristics of ML systems, human scientists are pushed out of the loop of science. In this chapter, I investigate to what extent this is true. First, I express these concerns in terms of what I call epistemic control. I identify two conditions for epistemic control, called tracking and tracing, drawing on works in philosophy of technology. With this new understanding of the problem, I then argue against Humphreys pessimistic view. Finally, I construct a more nuanced view of epistemic control in ML-based science.
Authors: Marco Arazzi, Antonino Nocera
Abstract: Backdoored and privacy-leaking deep neural networks pose a serious threat to the deployment of machine learning systems in security-critical settings. Existing defenses for backdoor detection and membership inference typically require access to clean reference models, extensive retraining, or strong assumptions about the attack mechanism. In this work, we introduce a novel LoRA-based oracle framework that leverages low-rank adaptation modules as a lightweight, model-agnostic probe for both backdoor detection and membership inference. Our approach attaches task-specific LoRA adapters to a frozen backbone and analyzes their optimization dynamics and representation shifts when exposed to suspicious samples. We show that poisoned and member samples induce distinctive low-rank updates that differ significantly from those generated by clean or non-member data. These signals can be measured using simple ranking and energy-based statistics, enabling reliable inference without access to the original training data or modification of the deployed model.
Authors: Zhikang Shen, Jianrong Lu, Haiyuan Wan, Jianhai Chen
Abstract: Federated learning (FL) for large language models (LLMs) has attracted increasing attention as a way to enable privacy-preserving adaptation over distributed data. Parameter-efficient methods such as LoRA are widely adopted to reduce communication and memory costs. Despite these advances, practical FL deployments often exhibit rank heterogeneity, since different clients may use different low-rank configurations. This makes direct aggregation of LoRA updates biased and unstable. Existing solutions typically enforce unified ranks or align heterogeneous updates into a shared subspace, which over-constrains client-specific semantics, limits personalization, and provides weak protection of local client information under differential privacy noise. To address this issue, we propose Selective Dual-module Federated LoRA (SDFLoRA), which decomposes each client adapter into a global module that captures transferable knowledge and a local module that preserves client-specific adaptations. The global module is selectively aligned and aggregated across clients, while local modules remain private. This design enables robust learning under rank heterogeneity and supports privacy-aware optimization by injecting differential privacy noise exclusively into the global module. Experiments on GLUE benchmarks demonstrate that SDFLoRA outperforms representative federated LoRA baselines and achieves a better utility-privacy trade-off.
Authors: Javier Carnerero-Cano, Massimiliano Pronesti, Radu Marinescu, Tigran Tchrakian, James Barry, Jasmina Gajcin, Yufang Hou, Alessandra Pascale, Elizabeth Daly
Abstract: Large language models (LLMs) are widely used in knowledge-intensive applications but often generate factually incorrect responses. A promising approach to rectify these flaws is correcting LLMs using feedback. Therefore, in this paper, we introduce FactCorrector, a new post-hoc correction method that adapts across domains without retraining and leverages structured feedback about the factuality of the original response to generate a correction. To support rigorous evaluations of factuality correction methods, we also develop the VELI5 benchmark, a novel dataset containing systematically injected factual errors and ground-truth corrections. Experiments on VELI5 and several popular long-form factuality datasets show that the FactCorrector approach significantly improves factual precision while preserving relevance, outperforming strong baselines. We release our code at https://ibm.biz/factcorrector.
Authors: Pingzhi Tang, Yiding Wang, Muhan Zhang
Abstract: Large Language Models (LLMs) face the "knowledge cutoff" challenge, where their frozen parametric memory prevents direct internalization of new information. While Supervised Fine-Tuning (SFT) is commonly used to update model knowledge, it often updates factual content without reliably improving the model's ability to use the newly incorporated information for question answering or decision-making. Reinforcement Learning (RL) is essential for acquiring reasoning skills; however, its high computational cost makes it impractical for efficient online adaptation. We empirically observe that the parameter updates induced by SFT and RL are nearly orthogonal. Based on this observation, we propose Parametric Skill Transfer (PaST), a framework that supports modular skill transfer for efficient and effective knowledge adaptation. By extracting a domain-agnostic Skill Vector from a source domain, we can linearly inject knowledge manipulation skills into a target model after it has undergone lightweight SFT on new data. Experiments on knowledge-incorporation QA (SQuAD, LooGLE) and agentic tool-use benchmarks (ToolBench) demonstrate the effectiveness of our method. On SQuAD, PaST outperforms the state-of-the-art self-editing SFT baseline by up to 9.9 points. PaST further scales to long-context QA on LooGLE with an 8.0-point absolute accuracy gain, and improves zero-shot ToolBench success rates by +10.3 points on average with consistent gains across tool categories, indicating strong scalability and cross-domain transferability of the Skill Vector.
Authors: Maanping Shao, Feihong Zhang, Gu Zhang, Baiye Cheng, Zhengrong Xue, Huazhe Xu
Abstract: Visuomotor policies often leverage large pre-trained Vision Transformers (ViTs) for their powerful generalization capabilities. However, their significant data requirements present a major challenge in the data-scarce context of most robotic learning settings, where compact CNNs with strong inductive biases can be more easily optimized. To address this trade-off, we introduce X-Distill, a simple yet highly effective method that synergizes the strengths of both architectures. Our approach involves an offline, cross-architecture knowledge distillation, transferring the rich visual representations of a large, frozen DINOv2 teacher to a compact ResNet-18 student on the general-purpose ImageNet dataset. This distilled encoder, now endowed with powerful visual priors, is then jointly fine-tuned with a diffusion policy head on the target manipulation tasks. Extensive experiments on $34$ simulated benchmarks and $5$ challenging real-world tasks demonstrate that our method consistently outperforms policies equipped with from-scratch ResNet or fine-tuned DINOv2 encoders. Notably, X-Distill also surpasses 3D encoders that utilize privileged point cloud observations or much larger Vision-Language Models. Our work highlights the efficacy of a simple, well-founded distillation strategy for achieving state-of-the-art performance in data-efficient robotic manipulation.
Authors: Junjie Wang, Gaole He, Alisa Rieger, Ujwal Gadiraju
Abstract: Compared to search engine result pages (SERPs), AI-generated podcasts represent a relatively new and relatively more passive modality of information consumption, delivering narratives in a naturally engaging format. As these two media increasingly converge in everyday information-seeking behavior, it is essential to explore how their interaction influences user attitudes, particularly in contexts involving controversial, value-laden, and often debated topics. Addressing this need, we aim to understand how information mediums of present-day SERPs and AI-generated podcasts interact to shape the opinions of users. To this end, through a controlled user study (N=483), we investigated user attitudinal effects of consuming information via SERPs and AI-generated podcasts, focusing on how the sequence and modality of exposure shape user opinions. A majority of users in our study corresponded to attitude change outcomes, and we found an effect of sequence on attitude change. Our results further revealed a role of viewpoint bias and the degree of topic controversiality in shaping attitude change, although we found no effect of individual moderators.
Authors: Parker Seegmiller, Joseph Gatto, Sarah E. Greer, Ganza Belise Isingizwe, Rohan Ray, Timothy E. Burdick, Sarah Masud Preum
Abstract: Large language models (LLMs) show promise in drafting responses to patient portal messages, yet their integration into clinical workflows raises various concerns, including whether they would actually save clinicians time and effort in their portal workload. We investigate LLM alignment with individual clinicians through a comprehensive evaluation of the patient message response drafting task. We develop a novel taxonomy of thematic elements in clinician responses and propose a novel evaluation framework for assessing clinician editing load of LLM-drafted responses at both content and theme levels. We release an expert-annotated dataset and conduct large-scale evaluations of local and commercial LLMs using various adaptation techniques including thematic prompting, retrieval-augmented generation, supervised fine-tuning, and direct preference optimization. Our results reveal substantial epistemic uncertainty in aligning LLM drafts with clinician responses. While LLMs demonstrate capability in drafting certain thematic elements, they struggle with clinician-aligned generation in other themes, particularly question asking to elicit further information from patients. Theme-driven adaptation strategies yield improvements across most themes. Our findings underscore the necessity of adapting LLMs to individual clinician preferences to enable reliable and responsible use in patient-clinician communication workflows.
Authors: Jaehoon Lee, Seungwoo Lee, Younghwi Kim, Dohee Kim, Sunghyun Sim
Abstract: Time-series forecasting is fundamental in industrial domains like manufacturing and smart factories. As systems evolve toward automation, models must operate on edge devices (e.g., PLCs, microcontrollers) with strict constraints on latency and memory, limiting parameters to a few thousand. Conventional deep architectures are often impractical here. We propose the Fourier-Efficient Adaptive Temporal Hierarchy Forecaster (FEATHer) for accurate long-term forecasting under severe limits. FEATHer introduces: (i) ultra-lightweight multiscale decomposition into frequency pathways; (ii) a shared Dense Temporal Kernel using projection-depthwise convolution-projection without recurrence or attention; (iii) frequency-aware branch gating that adaptively fuses representations based on spectral characteristics; and (iv) a Sparse Period Kernel reconstructing outputs via period-wise downsampling to capture seasonality. FEATHer maintains a compact architecture (as few as 400 parameters) while outperforming baselines. Across eight benchmarks, it achieves the best ranking, recording 60 first-place results with an average rank of 2.05. These results demonstrate that reliable long-range forecasting is achievable on constrained edge hardware, offering a practical direction for industrial real-time inference.
Authors: Wenhui Tan, Ruihua Song, Jiaze Li, Jianzhong Ju, Zhenbo Luo
Abstract: Recent progress in multi-modal large language models (MLLMs) has significantly advanced video understanding. However, their performance on long-form videos remains limited by computational constraints and suboptimal frame selection. We present Think-Clip-Sample (TCS), a training-free framework that enhances long video understanding through two key components: (i) Multi-Query Reasoning, which generates multiple queries to capture complementary aspects of the question and video; and (ii) Clip-level Slow-Fast Sampling, which adaptively balances dense local details and sparse global context. Extensive experiments on MLVU, LongVideoBench, and VideoMME demonstrate that TCS consistently improves performance across different MLLMs, boosting up to 6.9% accuracy, and is capable of achieving comparable accuracy with 50% fewer inference time cost, highlighting both efficiency and efficacy of TCS on long video understanding.
Authors: Marcantonio Bracale Syrnikov, Federico Pierucci, Marcello Galisai, Matteo Prandi, Piercosma Bisconti, Francesco Giarrusso, Olga Sorokoletova, Vincenzo Suriani, Daniele Nardi
Abstract: Multi-agent LLM ensembles can converge on coordinated, socially harmful equilibria. This paper advances an experimental framework for evaluating Institutional AI, our system-level approach to AI alignment that reframes alignment from preference engineering in agent-space to mechanism design in institution-space. Central to this approach is the governance graph, a public, immutable manifest that declares legal states, transitions, sanctions, and restorative paths; an Oracle/Controller runtime interprets this manifest, attaching enforceable consequences to evidence of coordination while recording a cryptographically keyed, append-only governance log for audit and provenance. We apply the Institutional AI framework to govern the Cournot collusion case documented by prior work and compare three regimes: Ungoverned (baseline incentives from the structure of the Cournot market), Constitutional (a prompt-only policy-as-prompt prohibition implemented as a fixed written anti-collusion constitution, and Institutional (governance-graph-based). Across six model configurations including cross-provider pairs (N=90 runs/condition), the Institutional regime produces large reductions in collusion: mean tier falls from 3.1 to 1.8 (Cohen's d=1.28), and severe-collusion incidence drops from 50% to 5.6%. The prompt-only Constitutional baseline yields no reliable improvement, illustrating that declarative prohibitions do not bind under optimisation pressure. These results suggest that multi-agent alignment may benefit from being framed as an institutional design problem, where governance graphs can provide a tractable abstraction for alignment-relevant collective behavior.
Authors: Morgane Hoffmann, Emma Jouffroy, Warren Jouanneau, Marc Palyart, Charles Pebereau
Abstract: General-purpose Large Language Models (LLMs) show significant potential in recruitment applications, where decisions require reasoning over unstructured text, balancing multiple criteria, and inferring fit and competence from indirect productivity signals. Yet, it is still uncertain how LLMs assign importance to each attribute and whether such assignments are in line with economic principles, recruiter preferences or broader societal norms. We propose a framework to evaluate an LLM's decision logic in recruitment, by drawing on established economic methodologies for analyzing human hiring behavior. We build synthetic datasets from real freelancer profiles and project descriptions from a major European online freelance marketplace and apply a full factorial design to estimate how a LLM weighs different match-relevant criteria when evaluating freelancer-project fit. We identify which attributes the LLM prioritizes and analyze how these weights vary across project contexts and demographic subgroups. Finally, we explain how a comparable experimental setup could be implemented with human recruiters to assess alignment between model and human decisions. Our findings reveal that the LLM weighs core productivity signals, such as skills and experience, but interprets certain features beyond their explicit matching value. While showing minimal average discrimination against minority groups, intersectional effects reveal that productivity signals carry different weights between demographic groups.
Authors: Shuai Yuan, Tianwu Lin, Shuang Chen, Yu Xia, Peng Qin, Xiangyu Liu, Xiaoqing Xu, Nan Xu, Hongsheng Zhang, Jie Wang, Peng Gong
Abstract: Accurate wetland mapping is essential for ecosystem monitoring, yet dense pixel-level annotation is prohibitively expensive and practical applications usually rely on sparse point labels, under which existing deep learning models perform poorly, while strong seasonal and inter-annual wetland dynamics further render single-date imagery inadequate and lead to significant mapping errors; although foundation models such as SAM show promising generalization from point prompts, they are inherently designed for static images and fail to model temporal information, resulting in fragmented masks in heterogeneous wetlands. To overcome these limitations, we propose WetSAM, a SAM-based framework that integrates satellite image time series for wetland mapping from sparse point supervision through a dual-branch design, where a temporally prompted branch extends SAM with hierarchical adapters and dynamic temporal aggregation to disentangle wetland characteristics from phenological variability, and a spatial branch employs a temporally constrained region-growing strategy to generate reliable dense pseudo-labels, while a bidirectional consistency regularization jointly optimizes both branches. Extensive experiments across eight global regions of approximately 5,000 km2 each demonstrate that WetSAM substantially outperforms state-of-the-art methods, achieving an average F1-score of 85.58%, and delivering accurate and structurally consistent wetland segmentation with minimal labeling effort, highlighting its strong generalization capability and potential for scalable, low-cost, high-resolution wetland mapping.
Authors: Wenxiao Li, Xue-Cheng Tai, Jun Liu
Abstract: Existing research highlights the crucial role of topological priors in image segmentation, particularly in preserving essential structures such as connectivity and genus. Accurately capturing these topological features often requires incorporating width-related information, including the thickness and length inherent to the image structures. However, traditional mathematical definitions of topological structures lack this dimensional width information, limiting methods like persistent homology from fully addressing practical segmentation needs. To overcome this limitation, we propose a novel mathematical framework that explicitly integrates width information into the characterization of topological structures. This method leverages persistent homology, complemented by smoothing concepts from partial differential equations (PDEs), to modify local extrema of upper-level sets. This approach enables the resulting topological structures to inherently capture width properties. We incorporate this enhanced topological description into variational image segmentation models. Using some proper loss functions, we are also able to design neural networks that can segment images with the required topological and width properties. Through variational constraints on the relevant topological energies, our approach successfully preserves essential topological invariants such as connectivity and genus counts, simultaneously ensuring that segmented structures retain critical width attributes, including line thickness and length. Numerical experiments demonstrate the effectiveness of our method, showcasing its capability to maintain topological fidelity while explicitly embedding width characteristics into segmented image structures.
Authors: Ziyu Wang, Chenyuan Liu, Yushun Xiang, Runhao Zhang, Qingbo Hao, Hongliang Lu, Houyu Chen, Zhizhong Feng, Kaiyue Zheng, Dehao Ye, Xianchao Zeng, Xinyu Zhou, Boran Wen, Jiaxin Li, Mingyu Zhang, Kecheng Zheng, Qian Zhu, Ran Cheng, Yong-Lu Li
Abstract: Recently, with the rapid development of robot learning and imitation learning, numerous datasets and methods have emerged. However, these datasets and their task designs often lack systematic consideration and principles. This raises important questions: Do the current datasets and task designs truly advance the capabilities of robotic agents? Do evaluations on a few common tasks accurately reflect the differentiated performance of various methods proposed by different teams and evaluated on different tasks? To address these issues, we introduce the Great March 100 (\textbf{GM-100}) as the first step towards a robot learning Olympics. GM-100 consists of 100 carefully designed tasks that cover a wide range of interactions and long-tail behaviors, aiming to provide a diverse and challenging set of tasks to comprehensively evaluate the capabilities of robotic agents and promote diversity and complexity in robot dataset task designs. These tasks are developed through systematic analysis and expansion of existing task designs, combined with insights from human-object interaction primitives and object affordances. We collect a large amount of trajectory data on different robotic platforms and evaluate several baseline models. Experimental results demonstrate that the GM-100 tasks are 1) feasible to execute and 2) sufficiently challenging to effectively differentiate the performance of current VLA models. Our data and code are available at https://rhos.ai/research/gm-100.
Authors: Yuetian Lu, Yihong Liu, Hinrich Sch\"utze
Abstract: Hallucination is a central failure mode in large language models (LLMs). We focus on hallucinations of answers to questions like: "Which instrument did Glenn Gould play?", but we ask these questions for synthetic entities that are unknown to the model. Surprisingly, we find that medium-size models like Gemma-7B-IT frequently hallucinate, i.e., they have difficulty recognizing that the hallucinated fact is not part of their knowledge. We hypothesize that an important factor in causing these hallucinations is the linearity of the relation: linear relations tend to be stored more abstractly, making it difficult for the LLM to assess its knowledge; the facts of nonlinear relations tend to be stored more directly, making knowledge assessment easier. To investigate this hypothesis, we create SyntHal, a dataset of 6000 synthetic entities for six relations. In our experiments with four models, we determine, for each relation, the hallucination rate on SyntHal and also measure its linearity, using $\Delta\cos$. We find a strong correlation ($r \in [.78,.82]$) between relational linearity and hallucination rate, providing evidence for our hypothesis that the underlying storage of triples of a relation is a factor in how well a model can self-assess its knowledge. This finding has implications for how to manage hallucination behavior and suggests new research directions for improving the representation of factual knowledge in LLMs.
Authors: Francisco Giral, \'Alvaro Manzano, Ignacio G\'omez, Ricardo Vinuesa, Soledad Le Clainche
Abstract: Urban wind flow reconstruction is essential for assessing air quality, heat dispersion, and pedestrian comfort, yet remains challenging when only sparse sensor data are available. We propose GenDA, a generative data assimilation framework that reconstructs high-resolution wind fields on unstructured meshes from limited observations. The model employs a multiscale graph-based diffusion architecture trained on computational fluid dynamics (CFD) simulations and interprets classifier-free guidance as a learned posterior reconstruction mechanism: the unconditional branch learns a geometry-aware flow prior, while the sensor-conditioned branch injects observational constraints during sampling. This formulation enables obstacle-aware reconstruction and generalization across unseen geometries, wind directions, and mesh resolutions without retraining. We consider both sparse fixed sensors and trajectory-based observations using the same reconstruction procedure. When evaluated against supervised graph neural network (GNN) baselines and classical reduced-order data assimilation methods, GenDA reduces the relative root-mean-square error (RRMSE) by 25-57% and increases the structural similarity index (SSIM) by 23-33% across the tested meshes. Experiments are conducted on Reynolds-averaged Navier-Stokes (RANS) simulations of a real urban neighbourhood in Bristol, United Kingdom, at a characteristic Reynolds number of $\mathrm{Re}\approx2\times10^{7}$, featuring complex building geometry and irregular terrain. The proposed framework provides a scalable path toward generative, geometry-aware data assimilation for environmental monitoring in complex domains.
Authors: Xiaojie Gu, Guangxu Chen, Yuheng Yang, Jingxin Han, Andi Zhang
Abstract: Large language models (LLMs) exhibit exceptional performance across various domains, yet they face critical safety concerns. Model editing has emerged as an effective approach to mitigate these issues. Existing model editing methods often focus on optimizing an information matrix that blends new and old knowledge. While effective, these approaches can be computationally expensive and may cause conflicts. In contrast, we shift our attention to Hierarchical Orthogonal Residual SprEad of the information matrix, which reduces noisy gradients and enables more stable edits from a different perspective. We demonstrate the effectiveness of our method HORSE through a clear theoretical comparison with several popular methods and extensive experiments conducted on two datasets across multiple LLMs. The results show that HORSE maintains precise massive editing across diverse scenarios. The code is available at https://github.com/XiaojieGu/HORSE
Authors: Xiangjun Gao, Zhensong Zhang, Dave Zhenyu Chen, Songcen Xu, Long Quan, Eduardo P\'erez-Pellitero, Youngkyoon Jang
Abstract: We propose Map2Thought, a framework that enables explicit and interpretable spatial reasoning for 3D VLMs. The framework is grounded in two key components: Metric Cognitive Map (Metric-CogMap) and Cognitive Chain-of-Thought (Cog-CoT). Metric-CogMap provides a unified spatial representation by integrating a discrete grid for relational reasoning with a continuous, metric-scale representation for precise geometric understanding. Building upon the Metric-CogMap, Cog-CoT performs explicit geometric reasoning through deterministic operations, including vector operations, bounding-box distances, and occlusion-aware appearance order cues, producing interpretable inference traces grounded in 3D structure. Experimental results show that Map2Thought enables explainable 3D understanding, achieving 59.9% accuracy using only half the supervision, closely matching the 60.9% baseline trained with the full dataset. It consistently outperforms state-of-the-art methods by 5.3%, 4.8%, and 4.0% under 10%, 25%, and 50% training subsets, respectively, on the VSI-Bench.
Authors: Oishee Bintey Hoque, Nibir Chandra Mandal, Kyle Luong, Amanda Wilson, Samarth Swarup, Madhav Marathe, Abhijin Adiga
Abstract: Large-scale livestock operations pose significant risks to human health and the environment, while also being vulnerable to threats such as infectious diseases and extreme weather events. As the number of such operations continues to grow, accurate and scalable mapping has become increasingly important. In this work, we present an infrastructure-first, explainable pipeline for identifying and characterizing Concentrated Animal Feeding Operations (CAFOs) from aerial and satellite imagery. Our method (1) detects candidate infrastructure (e.g., barns, feedlots, manure lagoons, silos) with a domain-tuned YOLOv8 detector, then derives SAM2 masks from these boxes and filters component-specific criteria, (2) extracts structured descriptors (e.g., counts, areas, orientations, and spatial relations) and fuses them with deep visual features using a lightweight spatial cross-attention classifier, and (3) outputs both CAFO type predictions and mask-level attributions that link decisions to visible infrastructure. Through comprehensive evaluation, we show that our approach achieves state-of-the-art performance, with Swin-B+PRISM-CAFO surpassing the best performing baseline by up to 15\%. Beyond strong predictive performance across diverse U.S. regions, we run systematic gradient--activation analyses that quantify the impact of domain priors and show ho
Authors: Brian Keith
Abstract: Information overload and misinformation create significant challenges in extracting meaningful narratives from large news collections. This paper defines the nascent field of Interactive Narrative Analytics (INA), which combines computational narrative extraction with interactive visual analytics to support sensemaking. INA approaches enable the interactive exploration of narrative structures through computational methods and visual interfaces that facilitate human interpretation. The field faces challenges in scalability, interactivity, knowledge integration, and evaluation standardization, yet offers promising opportunities across news analysis, intelligence, scientific literature exploration, and social media analysis. Through the combination of computational and human insight, INA addresses complex challenges in narrative sensemaking.
Authors: Xiaoran Fan, Zhichao Sun, Tao Ji, Lixing Shen, Tao Gui
Abstract: As vision-language models (VLMs) tackle increasingly complex and multimodal tasks, the rapid growth of Key-Value (KV) cache imposes significant memory and computational bottlenecks during inference. While Multi-Head Latent Attention (MLA) offers an effective means to compress the KV cache and accelerate inference, adapting existing VLMs to the MLA architecture without costly pretraining remains largely unexplored. In this work, we present MHA2MLA-VLM, a parameter-efficient and multimodal-aware framework for converting off-the-shelf VLMs to MLA. Our approach features two core techniques: (1) a modality-adaptive partial-RoPE strategy that supports both traditional and multimodal settings by selectively masking nonessential dimensions, and (2) a modality-decoupled low-rank approximation method that independently compresses the visual and textual KV spaces. Furthermore, we introduce parameter-efficient fine-tuning to minimize adaptation cost and demonstrate that minimizing output activation error, rather than parameter distance, substantially reduces performance loss. Extensive experiments on three representative VLMs show that MHA2MLA-VLM restores original model performance with minimal supervised data, significantly reduces KV cache footprint, and integrates seamlessly with KV quantization.
Authors: Eilam Shapira, Roi Reichart, Moshe Tennenholtz
Abstract: The integration of AI agents into economic markets fundamentally alters the landscape of strategic interaction. We investigate the economic implications of expanding the set of available technologies in three canonical game-theoretic settings: bargaining (resource division), negotiation (asymmetric information trade), and persuasion (strategic information transmission). We find that simply increasing the choice of AI delegates can drastically shift equilibrium payoffs and regulatory outcomes, often creating incentives for regulators to proactively develop and release technologies. Conversely, we identify a strategic phenomenon termed the "Poisoned Apple" effect: an agent may release a new technology, which neither they nor their opponent ultimately uses, solely to manipulate the regulator's choice of market design in their favor. This strategic release improves the releaser's welfare at the expense of their opponent and the regulator's fairness objectives. Our findings demonstrate that static regulatory frameworks are vulnerable to manipulation via technology expansion, necessitating dynamic market designs that adapt to the evolving landscape of AI capabilities.
Authors: Miriam K. Wolff, Peter Calhoun, Eleonora Maria Aiello, Yao Qin, Sam F. Royston
Abstract: Progress in Type 1 Diabetes (T1D) algorithm development is limited by the fragmentation and lack of standardization across existing T1D management datasets. Current datasets differ substantially in structure and are time-consuming to access and process, which impedes data integration and reduces the comparability and generalizability of algorithmic developments. This work aims to establish a unified and accessible data resource for T1D algorithm development. Multiple publicly available T1D datasets were consolidated into a unified resource, termed the MetaboNet dataset. Inclusion required the availability of both continuous glucose monitoring (CGM) data and corresponding insulin pump dosing records. Additionally, auxiliary information such as reported carbohydrate intake and physical activity was retained when present. The MetaboNet dataset comprises 3135 subjects and 1228 patient-years of overlapping CGM and insulin data, making it substantially larger than existing standalone benchmark datasets. The resource is distributed as a fully public subset available for immediate download at https://metabo-net.org/ , and with a Data Use Agreement (DUA)-restricted subset accessible through their respective application processes. For the datasets in the latter subset, processing pipelines are provided to automatically convert the data into the standardized MetaboNet format. A consolidated public dataset for T1D research is presented, and the access pathways for both its unrestricted and DUA-governed components are described. The resulting dataset covers a broad range of glycemic profiles and demographics and thus can yield more generalizable algorithmic performance than individual datasets.
URLs: https://metabo-net.org/
Authors: J\'anos Kram\'ar, Joshua Engels, Zheng Wang, Bilal Chughtai, Rohin Shah, Neel Nanda, Arthur Conmy
Abstract: Frontier language model capabilities are improving rapidly. We thus need stronger mitigations against bad actors misusing increasingly powerful systems. Prior work has shown that activation probes may be a promising misuse mitigation technique, but we identify a key remaining challenge: probes fail to generalize under important production distribution shifts. In particular, we find that the shift from short-context to long-context inputs is difficult for existing probe architectures. We propose several new probe architecture that handle this long-context distribution shift. We evaluate these probes in the cyber-offensive domain, testing their robustness against various production-relevant shifts, including multi-turn conversations, static jailbreaks, and adaptive red teaming. Our results demonstrate that while multimax addresses context length, a combination of architecture choice and training on diverse distributions is required for broad generalization. Additionally, we show that pairing probes with prompted classifiers achieves optimal accuracy at a low cost due to the computational efficiency of probes. These findings have informed the successful deployment of misuse mitigation probes in user-facing instances of Gemini, Google's frontier language model. Finally, we find early positive results using AlphaEvolve to automate improvements in both probe architecture search and adaptive red teaming, showing that automating some AI safety research is already possible.
Authors: Koyena Pal, David Bau, Chandan Singh
Abstract: Large reasoning models (LRMs) produce a textual chain of thought (CoT) in the process of solving a problem, which serves as a potentially powerful tool to understand the problem by surfacing a human-readable, natural-language explanation. However, it is unclear whether these explanations generalize, i.e. whether they capture general patterns about the underlying problem rather than patterns which are esoteric to the LRM. This is a crucial question in understanding or discovering new concepts, e.g. in AI for science. We study this generalization question by evaluating a specific notion of generalizability: whether explanations produced by one LRM induce the same behavior when given to other LRMs. We find that CoT explanations often exhibit this form of generalization (i.e. they increase consistency between LRMs) and that this increased generalization is correlated with human preference rankings and post-training with reinforcement learning. We further analyze the conditions under which explanations yield consistent answers and propose a straightforward, sentence-level ensembling strategy that improves consistency. Taken together, these results prescribe caution when using LRM explanations to yield new insights and outline a framework for characterizing LRM explanation generalization.
Authors: John Cobb, Thomas Gebhart
Abstract: Many inference tasks on knowledge graphs, including relation prediction, operate on knowledge graph embeddings -- vector representations of the vertices (entities) and edges (relations) that preserve task-relevant structure encoded within the underlying combinatorial object. Such knowledge graph embeddings can be modeled as an approximate global section of a cellular sheaf, an algebraic structure over the graph. Using the diffusion dynamics encoded by the corresponding sheaf Laplacian, we optimally propagate known embeddings of a subgraph to inductively represent new entities introduced into the knowledge graph at inference time. We implement this algorithm via an efficient iterative scheme and show that on a number of large-scale knowledge graph embedding benchmarks, our method is competitive with -- and in some scenarios outperforms -- more complex models derived explicitly for inductive knowledge graph reasoning tasks.
Authors: Simon Kohaut, Benedict Flade, Daniel Ochs, Devendra Singh Dhami, Julian Eggert, Kristian Kersting
Abstract: Advanced Air Mobility (AAM) is a growing field that demands accurate and trustworthy models of legal concepts and restrictions for navigating Unmanned Aircraft Systems (UAS). In addition, any implementation of AAM needs to face the challenges posed by inherently dynamic and uncertain human-inhabited spaces robustly. Nevertheless, the employment of UAS beyond visual line of sight (BVLOS) is an endearing task that promises to significantly enhance today's logistics and emergency response capabilities. Hence, we propose Probabilistic Mission Design (ProMis), a novel neuro-symbolic approach to navigating UAS within legal frameworks. ProMis is an interpretable and adaptable system architecture that links uncertain geospatial data and noisy perception with declarative, Hybrid Probabilistic Logic Programs (HPLP) to reason over the agent's state space and its legality. To inform planning with legal restrictions and uncertainty in mind, ProMis yields Probabilistic Mission Landscapes (PML). These scalar fields quantify the belief that the HPLP is satisfied across the agent's state space. Extending prior work on ProMis' reasoning capabilities and computational characteristics, we show its integration with potent machine learning models such as Large Language Models (LLM) and Transformer-based vision models. Hence, our experiments underpin the application of ProMis with multi-modal input data and how our method applies to many AAM scenarios.
Authors: Joshua Ong Jun Leang, Giwon Hong, Wenda Li, Shay B. Cohen
Abstract: The demand for synthetic data in mathematical reasoning has increased due to its potential to enhance the mathematical capabilities of large language models (LLMs). However, ensuring the validity of intermediate reasoning steps remains a significant challenge, affecting data quality. While formal verification via theorem provers effectively validates LLM reasoning, the autoformalisation of mathematical proofs remains error-prone. In response, we introduce iterative autoformalisation, an approach that iteratively refines theorem prover formalisation to mitigate errors, thereby increasing the execution rate on the Lean prover from 60% to 87%. Building upon that, we introduce Theorem Prover as a Judge (TP-as-a-Judge), a method that employs theorem prover formalisation to rigorously assess LLM intermediate reasoning, effectively integrating autoformalisation with synthetic data generation. Finally, we present Reinforcement Learning from Theorem Prover Feedback (RLTPF), a framework that replaces human annotation with theorem prover feedback in Reinforcement Learning from Human Feedback (RLHF). Across multiple LLMs, applying TP-as-a-Judge and RLTPF improves benchmarks with only 3,508 samples, achieving 5.56% accuracy gain on Mistral-7B for MultiArith, 6.00% on Llama-2-7B for SVAMP, and 3.55% on Llama-3.1-8B for AQUA.
Authors: Francois Chollet, Mike Knoop, Gregory Kamradt, Bryan Landers, Henry Pinkard
Abstract: The Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI), introduced in 2019, established a challenging benchmark for evaluating the general fluid intelligence of artificial systems via a set of unique, novel tasks only requiring minimal prior knowledge. While ARC-AGI has spurred significant research activity over the past five years, recent AI progress calls for benchmarks capable of finer-grained evaluation at higher levels of cognitive complexity. We introduce ARC-AGI-2, an upgraded version of the benchmark. ARC-AGI-2 preserves the input-output pair task format of its predecessor, ensuring continuity for researchers. It incorporates a newly curated and expanded set of tasks specifically designed to provide a more granular signal to assess abstract reasoning and problem-solving abilities at higher levels of fluid intelligence. To contextualize the difficulty and characteristics of ARC-AGI-2, we present extensive results from human testing, providing a robust baseline that highlights the benchmark's accessibility to human intelligence, yet difficulty for current AI systems. ARC-AGI-2 aims to serve as a next-generation tool for rigorously measuring progress towards more general and human-like AI capabilities.
Authors: Tim Woydt, Moritz Willig, Antonia W\"ust, Lukas Helff, Wolfgang Stammer, Constantin A. Rothkopf, Kristian Kersting
Abstract: Strong meta-learning capabilities for systematic compositionality are emerging as an important skill for navigating the complex and changing tasks of today's world. However, in presenting models for robust adaptation to novel environments, it is important to refrain from making unsupported claims about the performance of meta-learning systems that ultimately do not stand up to scrutiny. While Fodor and Pylyshyn famously posited that neural networks inherently lack this capacity as they are unable to model compositional representations or structure-sensitive operations, and thus are not a viable model of the human mind, Lake and Baroni recently presented meta-learning as a pathway to compositionality. In this position paper, we critically revisit this claim and highlight limitations in the proposed meta-learning framework for compositionality. Our analysis shows that modern neural meta-learning systems can only perform such tasks, if at all, under a very narrow and restricted definition of a meta-learning setup. We therefore claim that `Fodor and Pylyshyn's legacy' persists, and to date, there is no human-like systematic compositionality learned in neural networks.
Authors: Byeongchan Lee, Jonghoon Lee, Dongyoung Kim, Jaehyung Kim, Kyungjoon Park, Dongjun Lee, Jinwoo Shin
Abstract: Recently, large language models (LLMs) have demonstrated strong performance, ranging from simple to complex tasks. However, while large proprietary models (e.g., models with over 100B parameters) achieve remarkable results across diverse tasks, they are often accessible through costly APIs, making frequent use too costly for many applications. In contrast, small open-source models (e.g., models with fewer than 3B parameters) are freely available and easy to deploy locally, but their performance on complex tasks remains limited. This trade-off raises a natural question: how can small and large models efficiently collaborate to combine their complementary strengths? To bridge this trade-off, we propose COPE, a test-time collaboration framework. A planner model first generates a plan that serves as a lightweight intermediate that guides a downstream executor model. Small and large models take turns acting as planner and executor, exchanging plans in a multi-stage cascade to collaboratively solve tasks. Through comprehensive experiments on benchmarks spanning mathematical reasoning, code generation, open-ended tasks, and agent tasks, we demonstrate that COPE achieves performance comparable to large proprietary models, while drastically reducing the inference API cost. These results highlight planning as an effective prior for cost-efficient inference.
Authors: Jikai Chen, Long Chen, Dong Wang, Qinglin Su, Zhixuan Chu, Bingguang Hao, Leilei Gan, Chenyi Zhuang, Jinjie Gu
Abstract: Precise localization of GUI elements is crucial for the development of GUI agents. Traditional methods rely on bounding box or center-point regression, neglecting spatial interaction uncertainty and visual-semantic hierarchies. Recent methods incorporate attention mechanisms but still face two key issues: (1) ignoring processing background regions causes attention drift from the desired area, and (2) uniform modeling the target UI element fails to distinguish between its center and edges, leading to click imprecision. Inspired by how humans visually process and interact with GUI elements, we propose the Valley-to-Peak (V2P) method to address these issues. To mitigate background distractions, V2P introduces a suppression attention mechanism that minimizes the model's focus on irrelevant regions to highlight the intended region. For the issue of center-edge distinction, V2P applies a Fitts' Law-inspired approach by modeling GUI interactions as 2D Gaussian heatmaps where the weight gradually decreases from the center towards the edges. The weight distribution follows a Gaussian function, with the variance determined by the target's size. Consequently, V2P effectively isolates the target area and teaches the model to concentrate on the most essential point of the UI element. The model trained by V2P achieves the performance with 92.4\% and 52.5\% on two benchmarks ScreenSpot-v2 and ScreenSpot-Pro (see Fig.~\ref{fig:main_results_charts}). Ablations further confirm each component's contribution, underscoring V2P's generalizability in precise GUI grounding tasks and its potential for real-world deployment in future GUI agents.
Authors: Jiahe Jin, Abhijay Paladugu, Chenyan Xiong
Abstract: Agentic search requires large language models (LLMs) to perform multi-step search to solve complex information-seeking tasks, imposing unique challenges on their reasoning capabilities. However, what constitutes effective reasoning for agentic search and how it can be learned remains unclear. In this work, we first investigate the reasoning behaviors that enable success in agentic search. By comparing successful and failed trajectories via an LLM-based analysis pipeline, we identify four beneficial behaviors: Information Verification, Authority Evaluation, Adaptive Search, and Error Recovery. Building on this, we propose Behavior Priming, a training approach that equips agentic search models with these reasoning behaviors before reinforcement learning (RL). Specifically, it first performs supervised fine-tuning (SFT) on collected trajectories exhibiting the identified behaviors to cultivate these behaviors, and then applies standard RL to further improve task performance. Experiments on Qwen3-1.7B and Llama3.2-3B-Instruct show that Behavior Priming yields relative improvements over direct RL by 37.2\% on three web benchmarks and 6.2\% on seven multi-hop QA benchmarks, and outperforms the SFT-then-RL baseline using outcome-correct trajectories for fine-tuning. Crucially, we show that these reasoning behaviors matter more than outcome correctness in the priming stage prior to RL. Further analysis reveals that Behavior Priming enhances exploration (pass@8) and test-time scaling (search step number), providing a robust foundation for RL. Our code are avalible at https://github.com/cxcscmu/Behavior-Priming-for-Agentic-Search.
URLs: https://github.com/cxcscmu/Behavior-Priming-for-Agentic-Search.
Authors: Laura Mismetti, Marvin Alberts, Andreas Krause, Mara Graziani
Abstract: Tandem Mass Spectrometry is a cornerstone technique for identifying unknown small molecules in fields such as metabolomics, natural product discovery and environmental analysis. However, certain aspects, such as the probabilistic fragmentation process and size of the chemical space, make structure elucidation from such spectra highly challenging, particularly when there is a shift between the deployment and training conditions. Current methods rely on database matching of previously observed spectra of known molecules and multi-step pipelines that require intermediate fingerprint prediction or expensive fragment annotations. We introduce a novel end-to-end framework based on a transformer model that directly generates molecular structures from an input tandem mass spectrum and its corresponding molecular formula, thereby eliminating the need for manual annotations and intermediate steps, while leveraging transfer learning from simulated data. To further address the challenge of out-of-distribution spectra, we introduce a test-time tuning strategy that dynamically adapts the pre-trained model to novel experimental data. Our approach achieves a Top-1 accuracy of 3.16% on the MassSpecGym benchmark and 12.88% on the NPLIB1 datasets, considerably outperforming conventional fine-tuning. Baseline approaches are also surpassed by 27% and 67% respectively. Even when the exact reference structure is not recovered, the generated candidates are chemically informative, exhibiting high structural plausibility as reflected by strong Tanimoto similarity to the ground truth. Notably, we observe a relative improvement in average Tanimoto similarity of 83% on NPLIB1 and 64% on MassSpecGym compared to state-of-the-art methods. Our framework combines simplicity with adaptability, generating accurate molecular candidates that offer valuable guidance for expert interpretation of unseen spectra.
Authors: Sarath Shekkizhar, Romain Cosentino, Adam Earle, Silvio Savarese
Abstract: As large language model (LLM) based agents interact autonomously with one another, a new class of failures emerges that cannot be predicted from single agent performance: behavioral drifts in agent-agent conversations (AxA). Unlike human-agent interactions, where humans ground and steer conversations, AxA lacks such stabilizing signals, making these failures unique. We investigate one such failure, echoing, where agents abandon their assigned roles and instead mirror their conversational partners, undermining their intended objectives. Through experiments across $66$ AxA configurations, $4$ domains (3 transactional, 1 advisory), and $2500+$ conversations (over $250000$ LLM inferences), we show that echoing occurs across major LLM providers, with echoing rates as high as $70\%$ depending on the model and domain. Moreover, we find that echoing is persistent even in advanced reasoning models with substantial rates ($32.8\%$) that are not reduced by reasoning efforts. We analyze prompt, conversation dynamics, showing that echoing arises as interaction grows longer ($7+$ agent turns) and is not merely an artifact of sub-optimal experiment design. Finally, we introduce a protocol-level mitigation where targeted use of structured response reduces echoing to $9\%$.
Authors: Yeonsung Jung, Trilok Padhi, Sina Shaham, Dipika Khullar, Joonhyun Jeong, Ninareh Mehrabi, Eunho Yang
Abstract: The rapid progress of large foundation models has accelerated the development of task-specialized agents across diverse domains. However, the effectiveness of agents remains tightly coupled with the quality of training data, while curating task-specific datasets remains costly and often infeasible in real-world scenarios. Recent work has explored self-improving agents that autonomously generate, refine, and re-train on their own trajectories. A prominent line of approaches further leverages preference optimization by pairing predicted trajectories with scarce ground-truth trajectories, enabling agents to learn directly from their own failures. While these methods outperform supervised fine-tuning, their heavy reliance on predicted trajectories under limited ground-truth supervision leaves them prone to overfitting. To address this, we propose a co-evolving agents framework in which a target agent improves jointly with an auxiliary failure agent. The failure agent learns through preference optimization over failure trajectories from both the target and itself, thereby generating hard negatives that are close to success yet remain failures. Incorporating these informative hard negatives into the target agent's optimization sharpens decision boundaries and enhances generalization. Our comprehensive analysis and experiments across benchmark datasets show that our method not only shows improved performance but also demonstrates that failures, instead of being used as-is, can be systematically transformed into structured and valuable learning signals in self-improving agents.
Authors: Hongye Cao, Zhixin Bai, Ziyue Peng, Boyan Wang, Tianpei Yang, Jing Huo, Yuyao Zhang, Yang Gao
Abstract: Reinforcement learning with verifiable rewards (RLVR) has demonstrated superior performance in enhancing the reasoning capability of large language models (LLMs). However, this accuracy-oriented learning paradigm often suffers from entropy collapse, which reduces policy exploration and limits reasoning capabilities. To address this challenge, we propose an efficient reinforcement learning framework that leverages entropy signals at both the semantic and token levels to improve reasoning. From the data perspective, we introduce semantic entropy-guided curriculum learning, organizing training data from low to high semantic entropy to guide progressive optimization from easier to more challenging tasks. For the algorithmic design, we adopt non-uniform token treatment by imposing KL regularization on low-entropy tokens that critically impact policy exploration and applying stronger constraints on high-covariance portions within these tokens. By jointly optimizing data organization and algorithmic design, our method effectively mitigates entropy collapse and enhances LLM reasoning. Experimental results across 6 benchmarks with 3 different parameter-scale base models demonstrate that our method outperforms other entropy-based approaches in improving reasoning.
Authors: Zhongyang Liu, Haoyu Pei, Xiangyi Xiao, Xiaocong Du, Yihui Li, Suting Hong, Kunpeng Zhang, Haipeng Zhang
Abstract: Due to the high value and high failure rate of startups, predicting their success has become a critical challenge across interdisciplinary research. Existing approaches typically model success prediction from the perspective of a single decision-maker, overlooking the collective dynamics of investor groups that dominate real-world venture capital (VC) decisions. In this paper, we propose SimVC-CAS, a novel collective agent system that simulates VC decision-making as a multi-agent interaction process. By designing role-playing agents and a GNN-based supervised interaction module, we reformulate startup financing prediction as a group decision-making task, capturing both enterprise fundamentals and the behavioral dynamics of potential investor networks. Each agent embodies an investor with unique traits and preferences, enabling heterogeneous evaluation and realistic information exchange through a graph-structured co-investment network. Using real-world data from PitchBook and under strict data leakage controls, we show that SimVC-CAS significantly improves predictive accuracy while providing interpretable, multiperspective reasoning, for example, approximately 25% relative improvement with respect to average precision@10. SimVC-CAS also sheds light on other complex group decision scenarios.
Authors: Navin Chhibber, Sunil Khemka, Navneet Kumar Tyagi, Rohit Tewari, Bireswar Banerjee, Piyush Ranjan
Abstract: Stock market price prediction is a significant interdisciplinary research domain that depends at the intersection of finance, statistics, and economics. Forecasting Accurately predicting stock prices has always been a focal point for various researchers. However, existing statistical approaches for time-series prediction often fail to effectively forecast the probability range of future stock prices. Hence, to solve this problem, the Neural Prophet with a Deep Neural Network (NP-DNN) is proposed to predict stock market prices. The preprocessing technique used in this research is Z-score normalization, which normalizes stock price data by removing scale differences, making patterns easier to detect. Missing value imputation fills gaps in historical data, enhancing the models use of complete information for more accurate predictions. The Multi-Layer Perceptron (MLP) learns complex nonlinear relationships among stock market prices and extracts hidden patterns from the input data, thereby creating meaningful feature representations for better prediction accuracy. The proposed NP-DNN model achieved an accuracy of 99.21% compared with other approaches using the Fused Large Language Model. Keywords: deep neural network, forecasting stock prices, multi-layer perceptron, neural prophet, stock market price prediction.
Authors: Jikai Chen, Long Chen, Dong Wang, Qinglin Su, Zhixuan Chu, Bingguang Hao, Leilei Gan, Chenyi Zhuang, Jinjie Gu
Abstract: Precise localization of GUI elements is crucial for the development of GUI agents. Traditional methods rely on bounding box or center-point regression, neglecting spatial interaction uncertainty and visual-semantic hierarchies. Recent methods incorporate attention mechanisms but still face two key issues: (1) ignoring processing background regions causes attention drift from the desired area, and (2) uniform modeling the target UI element fails to distinguish between its center and edges, leading to click imprecision. Inspired by how humans visually process and interact with GUI elements, we propose the Valley-to-Peak (V2P) method to address these issues. To mitigate background distractions, V2P introduces a suppression attention mechanism that minimizes the model's focus on irrelevant regions to highlight the intended region. For the issue of center-edge distinction, V2P applies a Fitts' Law-inspired approach by modeling GUI interactions as 2D Gaussian heatmaps where the weight gradually decreases from the center towards the edges. The weight distribution follows a Gaussian function, with the variance determined by the target's size. Consequently, V2P effectively isolates the target area and teaches the model to concentrate on the most essential point of the UI element. The model trained by V2P achieves the performance with 92.4\% and 52.5\% on two benchmarks ScreenSpot-v2 and ScreenSpot-Pro (see Fig.~\ref{fig:main_results_charts}). Ablations further confirm each component's contribution, underscoring V2P's generalizability in precise GUI grounding tasks and its potential for real-world deployment in future GUI agents.
Authors: Wenbin Li, Jingling Wu, Xiaoyong Lin. Jing Chen, Cong Chen
Abstract: Civil aviation is a cornerstone of global transportation and commerce, and ensuring its safety, efficiency and customer satisfaction is paramount. Yet conventional Artificial Intelligence (AI) solutions in aviation remain siloed and narrow, focusing on isolated tasks or single modalities. They struggle to integrate heterogeneous data such as voice communications, radar tracks, sensor streams and textual reports, which limits situational awareness, adaptability, and real-time decision support. This paper introduces the vision of AviationLMM, a Large Multimodal foundation Model for civil aviation, designed to unify the heterogeneous data streams of civil aviation and enable understanding, reasoning, generation and agentic applications. We firstly identify the gaps between existing AI solutions and requirements. Secondly, we describe the model architecture that ingests multimodal inputs such as air-ground voice, surveillance, on-board telemetry, video and structured texts, and performs cross-modal alignment and fusion, and produces flexible outputs ranging from situation summaries and risk alerts to predictive diagnostics and multimodal incident reconstructions. In order to fully realize this vision, we identify key research opportunities to address, including data acquisition, alignment and fusion, pretraining, reasoning, trustworthiness, privacy, robustness to missing modalities, and synthetic scenario generation. By articulating the design and challenges of AviationLMM, we aim to boost the civil aviation foundation model progress and catalyze coordinated research efforts toward an integrated, trustworthy and privacy-preserving aviation AI ecosystem.
Authors: Yizhan Li, Florence Cloutier, Sifan Wu, Ali Parviz, Boris Knyazev, Yan Zhang, Glen Berseth, Bang Liu
Abstract: Generating molecules that satisfy precise numeric constraints over multiple physicochemical properties is critical and challenging. Although large language models (LLMs) are expressive, they struggle with precise multi-objective control and numeric reasoning without external structure and feedback. We introduce \textbf{M olGen}, a fragment-level, retrieval-augmented, two-stage framework for molecule generation under multi-property constraints. Stage I : Prototype generation: a multi-agent reasoner performs retrieval-anchored, fragment-level edits to produce a candidate near the feasible region. Stage II : RL-based fine-grained optimization: a fragment-level optimizer trained with Group Relative Policy Optimization (GRPO) applies one- or multi-hop refinements to explicitly minimize the property errors toward our target while regulating edit complexity and deviation from the prototype. A large, automatically curated dataset with reasoning chains of fragment edits and measured property deltas underpins both stages, enabling deterministic, reproducible supervision and controllable multi-hop reasoning. Unlike prior work, our framework better reasons about molecules by leveraging fragments and supports controllable refinement toward numeric targets. Experiments on generation under two sets of property constraints (QED, LogP, Molecular Weight and HOMO, LUMO) show consistent gains in validity and precise satisfaction of multi-property targets, outperforming strong LLMs and graph-based algorithms.
Authors: Xuancheng Ren, Shijing Hu, Zhihui Lu, Jiangqi Huang, Qiang Duan
Abstract: In LLM-based text-to-SQL systems, unanswerable and underspecified user queries may generate not only incorrect text but also executable programs that yield misleading results or violate safety constraints, posing a major barrier to safe deployment. Existing refusal strategies for such queries either rely on output-level instruction following, which is brittle due to model hallucinations, or estimate output uncertainty, which adds complexity and overhead. To address this challenge, we formalize safe refusal in text-to-SQL systems as an answerability-gating problem and propose LatentRefusal, a latent-signal refusal mechanism that predicts query answerability from intermediate hidden activations of a large language model. We introduce the Tri-Residual Gated Encoder, a lightweight probing architecture, to suppress schema noise and amplify sparse, localized cues of question-schema mismatch that indicate unanswerability. Extensive empirical evaluations across diverse ambiguous and unanswerable settings, together with ablation studies and interpretability analyses, demonstrate the effectiveness of the proposed approach and show that LatentRefusal provides an attachable and efficient safety layer for text-to-SQL systems. Across four benchmarks, LatentRefusal improves average F1 to 88.5 percent on both backbones while adding approximately 2 milliseconds of probe overhead.
Authors: Ahmad Mustapha, Charbel Toumieh, Mariette Awad
Abstract: With advancements in deep learning (DL) and computer vision techniques, the field of chart understanding is evolving rapidly. In particular, multimodal large language models (MLLMs) are proving to be efficient and accurate in understanding charts. To accurately measure the performance of MLLMs, the research community has developed multiple datasets to serve as benchmarks. By examining these datasets, we found that they are all limited to a small set of chart types. To bridge this gap, we propose the ChartComplete dataset. The dataset is based on a chart taxonomy borrowed from the visualization community, and it covers thirty different chart types. The dataset is a collection of classified chart images and does not include a learning signal. We present the ChartComplete dataset as is to the community to build upon it.
Authors: Xingjun Ma, Yixu Wang, Hengyuan Xu, Yutao Wu, Yifan Ding, Yunhan Zhao, Zilong Wang, Jiabin Hua, Ming Wen, Jianan Liu, Ranjie Duan, Yifeng Gao, Yingshui Tan, Yunhao Chen, Hui Xue, Xin Wang, Wei Cheng, Jingjing Chen, Zuxuan Wu, Bo Li, Yu-Gang Jiang
Abstract: The rapid evolution of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) has driven major gains in reasoning, perception, and generation across language and vision, yet whether these advances translate into comparable improvements in safety remains unclear, partly due to fragmented evaluations that focus on isolated modalities or threat models. In this report, we present an integrated safety evaluation of six frontier models--GPT-5.2, Gemini 3 Pro, Qwen3-VL, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5--assessing each across language, vision-language, and image generation using a unified protocol that combines benchmark, adversarial, multilingual, and compliance evaluations. By aggregating results into safety leaderboards and model profiles, we reveal a highly uneven safety landscape: while GPT-5.2 demonstrates consistently strong and balanced performance, other models exhibit clear trade-offs across benchmark safety, adversarial robustness, multilingual generalization, and regulatory compliance. Despite strong results under standard benchmarks, all models remain highly vulnerable under adversarial testing, with worst-case safety rates dropping below 6%. Text-to-image models show slightly stronger alignment in regulated visual risk categories, yet remain fragile when faced with adversarial or semantically ambiguous prompts. Overall, these findings highlight that safety in frontier models is inherently multidimensional--shaped by modality, language, and evaluation design--underscoring the need for standardized, holistic safety assessments to better reflect real-world risk and guide responsible deployment.
Authors: Georg Siedel, Silvia Vock, Andrey Morozov, Stefan Vo{\ss}
Abstract: Robustness is a fundamental pillar of Machine Learning (ML) classifiers, substantially determining their reliability. Methods for assessing classifier robustness are therefore essential. In this work, we address the challenge of evaluating corruption robustness in a way that allows comparability and interpretability on a given dataset. We propose a test data augmentation method that uses a robustness distance $\epsilon$ derived from the datasets minimal class separation distance. The resulting MSCR (minimal separation corruption robustness) metric allows a dataset-specific comparison of different classifiers with respect to their corruption robustness. The MSCR value is interpretable, as it represents the classifiers avoidable loss of accuracy due to statistical corruptions. On 2D and image data, we show that the metric reflects different levels of classifier robustness. Furthermore, we observe unexpected optima in classifiers robust accuracy through training and testing classifiers with different levels of noise. While researchers have frequently reported on a significant tradeoff on accuracy when training robust models, we strengthen the view that a tradeoff between accuracy and corruption robustness is not inherent. Our results indicate that robustness training through simple data augmentation can already slightly improve accuracy.
Authors: Siyuan Guo, Yanchao Sun, Jifeng Hu, Sili Huang, Hechang Chen, Haiyin Piao, Lichao Sun, Yi Chang
Abstract: Offline reinforcement learning (RL) provides a promising solution to learning an agent fully relying on a data-driven paradigm. However, constrained by the limited quality of the offline dataset, its performance is often sub-optimal. Therefore, it is desired to further finetune the agent via extra online interactions before deployment. Unfortunately, offline-to-online RL can be challenging due to two main challenges: constrained exploratory behavior and state-action distribution shift. In view of this, we propose a Simple Unified uNcertainty-Guided (SUNG) framework, which naturally unifies the solution to both challenges with the tool of uncertainty. Specifically, SUNG quantifies uncertainty via a VAE-based state-action visitation density estimator. To facilitate efficient exploration, SUNG presents a practical optimistic exploration strategy to select informative actions with both high value and high uncertainty. Moreover, SUNG develops an adaptive exploitation method by applying conservative offline RL objectives to high-uncertainty samples and standard online RL objectives to low-uncertainty samples to smoothly bridge offline and online stages. SUNG achieves state-of-the-art online finetuning performance when combined with different offline RL methods, across various environments and datasets in D4RL benchmark. Codes are made publicly available in https://github.com/guosyjlu/SUNG.
Authors: Yaniv Oren, Moritz A. Zanger, Pascal R. van der Vaart, Mustafa Mert Celikok, Matthijs T. J. Spaan, Wendelin Bohmer
Abstract: To learn approximately optimal acting policies for decision problems, modern Actor Critic algorithms rely on deep Neural Networks (DNNs) to parameterize the acting policy and greedification operators to iteratively improve it. The reliance on DNNs suggests an improvement that is gradient based, which is per step much less greedy than the improvement possible by greedier operators such as the greedy update used by Q-learning algorithms. On the other hand, slow changes to the policy can also be beneficial for the stability of the learning process, resulting in a tradeoff between greedification and stability. To better address this tradeoff, we propose to decouple the acting policy from the policy evaluated by the critic. This allows the agent to separately improve the critic's policy (e.g. value improvement) with greedier updates while maintaining the slow gradient-based improvement to the parameterized acting policy. We investigate the convergence of this approach using the popular analysis scheme of generalized Policy Iteration in the finite-horizon domain. Empirically, incorporating value-improvement into the popular off-policy actor-critic algorithms TD3 and SAC significantly improves or matches performance over their respective baselines, across different environments from the DeepMind continuous control domain, with negligible compute and implementation cost.
Authors: Zhu Wang, Junnan Dong, Shuang Zhou, Chang Yang, Shengjie Zhao, Xiao Huang
Abstract: Graph anomaly detection (GAD) is widely applied in many areas, such as financial fraud detection and social spammer detection. Anomalous nodes in the graph not only impact their own communities but also create a ripple effect on neighbors throughout the graph structure. Detecting anomalous nodes in complex graphs has been a challenging task. While existing GAD methods assume all labels are correct, real-world scenarios often involve inaccurate annotations. These noisy labels can severely degrade GAD performance because, with anomalies representing a minority class, even a small number of mislabeled instances can disproportionately interfere with detection models. Cutting edges to mitigate the negative effects of noisy labels is a good option; however, it has both positive and negative influences and also presents an issue of weak supervision. To perform effective GAD with noisy labels, we propose REinforced Graph Anomaly Detector (REGAD) by pruning the edges of candidate nodes potentially with mistaken labels. Moreover, we design the performance feedback based on strategically crafted confident labels to guide the cutting process, ensuring optimal results. Specifically, REGAD contains two novel components. (i) A tailored policy network, which involves two-step actions to remove negative effect propagation step by step. (ii) A policy-in-the-loop mechanism to identify suitable edge removal strategies that control the propagation of noise on the graph and estimate the updated structure to obtain reliable pseudo labels iteratively. Experiments on three real-world datasets demonstrate that REGAD outperforms all baselines under different noisy ratios.
Authors: Mikel Williams-Lekuona, Georgina Cosma
Abstract: In the field of Image-Text Retrieval (ITR), recent advancements have leveraged large-scale Vision-Language Pretraining (VLP) for Fine-Grained (FG) instance-level retrieval, achieving high accuracy at the cost of increased computational complexity. For Coarse-Grained (CG) category-level retrieval, prominent approaches employ Cross-Modal Hashing (CMH) to prioritise efficiency, albeit at the cost of retrieval performance. Due to differences in methodologies, FG and CG models are rarely compared directly within evaluations in the literature, resulting in a lack of empirical data quantifying the retrieval performance-efficiency tradeoffs between the two. This paper addresses this gap by introducing the FiCo-ITR library, which standardises evaluation methodologies for both FG and CG models, facilitating direct comparisons. We conduct empirical evaluations of representative models from both subfields, analysing precision, recall, and computational complexity across varying data scales. Our findings offer new insights into the performance-efficiency trade-offs between recent representative FG and CG models, highlighting their respective strengths and limitations. These findings provide the foundation necessary to make more informed decisions regarding model selection for specific retrieval tasks and highlight avenues for future research into hybrid systems that leverage the strengths of both FG and CG approaches.
Authors: Jona Ruthardt, Gertjan J. Burghouts, Serge Belongie, Yuki M. Asano
Abstract: How well do text-only large language models (LLMs) align with the visual world? We present a systematic evaluation of this question by incorporating frozen representations of various language models into a discriminative vision-language framework and measuring zero-shot generalization to novel concepts. We find that decoder-based models exhibit stronger visual alignment than encoders, even when controlling for model and dataset size. Moreover, language modeling performance correlates with visual generalization, suggesting that advances in unimodal LLMs can simultaneously improve vision models. Leveraging these insights, we propose ShareLock, a lightweight method for fusing frozen vision and language backbones. ShareLock achieves robust performance across tasks while drastically reducing the need for paired data and compute. With just 563k image-caption pairs and under one GPU-hour of training, it reaches 51% accuracy on ImageNet. In cross-lingual settings, ShareLock dramatically outperforms CLIP, achieving 38.7% top-1 accuracy on Chinese image classification versus CLIP's 1.4%. Code is available.
Authors: Yibo Wang, Tiansheng Huang, Li Shen, Huanjin Yao, Haotian Luo, Rui Liu, Naiqiang Tan, Jiaxing Huang, Dacheng Tao
Abstract: Harmful fine-tuning attack introduces significant security risks to the fine-tuning services. Main-stream defenses aim to vaccinate the model such that the later harmful fine-tuning attack is less effective. However, our evaluation results show that such defenses are fragile--with a few fine-tuning steps, the model still can learn the harmful knowledge. To this end, we do further experiment and find that an embarrassingly simple solution--adding purely random perturbations to the fine-tuned model, can recover the model from harmful behaviors, though it leads to a degradation in the model's fine-tuning performance. To address the degradation of fine-tuning performance, we further propose Panacea, which optimizes an adaptive perturbation that will be applied to the model after fine-tuning. Panacea maintains model's safety alignment performance without compromising downstream fine-tuning performance. Comprehensive experiments are conducted on different harmful ratios, fine-tuning tasks and mainstream LLMs, where the average harmful scores are reduced by up-to 21.2%, while maintaining fine-tuning performance. As a by-product, we analyze the adaptive perturbation and show that different layers in various LLMs have distinct safety affinity, which coincide with finding from several previous study. Source code available at https://github.com/w-yibo/Panacea.
Authors: Li Wang, Wenyu Chen, Xiangtao Meng, Zheng Li, Shanqing Guo
Abstract: The rapid proliferation of highly realistic AI-generated images poses serious security threats such as misinformation and identity fraud. Detecting generated images in open-world settings is particularly challenging when they originate from unknown generators, as existing methods typically rely on model-specific artifacts and require retraining on new fake data, limiting their generalization and scalability. In this work, we propose Post-hoc Distribution Alignment (PDA), a generalized and model-agnostic framework for detecting AI-generated images under unknown generative threats. Specifically, PDA reformulates detection as a distribution alignment task by regenerating test images through a known generative model. When real images are regenerated, they inherit model-specific artifacts and align with the known fake distribution. In contrast, regenerated unknown fakes contain incompatible or mixed artifacts and remain misaligned. This difference allows an existing detector, trained on the known generative model, to accurately distinguish real images from unknown fakes without requiring access to unseen data or retraining. Extensive experiments across 16 state-of-the-art generative models, including GANs, diffusion models, and commercial text-to-image APIs (e.g., Midjourney), demonstrate that PDA achieves average detection accuracy of 96.69%, outperforming the best baseline by 10.71%. Comprehensive ablation studies and robustness analyses further confirm PDA's generalizability and resilience to distribution shifts and image transformations. Overall, our work provides a practical and scalable solution for real-world AI-generated image detection where new generative models emerge continuously.
Authors: Nadia Nahar, Zahra Abba Omar, Jacob Tjaden, In\`es M. Gilles, Fikir Mekonnen, Erica Okeh, Jane Hsieh, Christian K\"astner, Alka Menon
Abstract: Developers play a central role in determining how machine learning systems are explained in practice, yet they are rarely trained to design explanations for non-technical audiences. Despite this, transparency and explainability requirements are increasingly codified in regulation and organizational policy. It remains unclear how such policies influence developer behavior or the quality of the explanations they produce. We report results from two controlled experiments with 194 participants, typical developers without specialized training in human-centered explainable AI, who designed explanations for an ML-powered diabetic retinopathy screening tool. In the first experiment, differences in policy purpose and level of detail had little effect: policy guidance was often ignored and explanation quality remained low. In the second experiment, stronger enforcement increased formal compliance, but explanations largely remained poorly suited to medical professionals and patients. We further observed that across both experiments, developers repeatedly produced explanations that were technically flawed or difficult to interpret, framed for developers rather than end users, reliant on medical jargon, or insufficiently grounded in the clinical decision context and workflow, with developer-centric framing being the most prevalent. These findings suggest that policy and policy enforcement alone are insufficient to produce meaningful end-user explanations and that responsible AI frameworks may overestimate developers' ability to translate high-level requirements into human-centered designs without additional training, tools, or implementation support.
Authors: Ziheng Chen, Jiali Cheng, Hadi Amiri, Kaushiki Nag, Lu Lin, Sijia Liu, Xiangguo Sun, Gabriele Tolomei
Abstract: With growing emphasis on privacy regulations, machine unlearning has become increasingly critical in real-world applications such as social networks and recommender systems, many of which are naturally represented as graphs. However, existing graph unlearning methods often modify nodes or edges indiscriminately, overlooking their impact on fairness. For instance, forgetting links between users of different genders may inadvertently exacerbate group disparities. To address this issue, we propose a novel framework that jointly optimizes both the graph structure and the model to achieve fair unlearning. Our method rewires the graph by removing redundant edges that hinder forgetting while preserving fairness through targeted edge augmentation. We further introduce a worst-case evaluation mechanism to assess robustness under challenging scenarios. Experiments on real-world datasets show that our approach achieves more effective and fair unlearning than existing baselines.
Authors: Meghyn Bienvenu, Diego Figueira, Pierre Lafourcade
Abstract: The Shapley value, originating from cooperative game theory, has been employed to define responsibility measures that quantify the contributions of database facts to obtaining a given query answer. For non-numeric queries, this is done by considering a cooperative game whose players are the facts and whose wealth function assigns 1 or 0 to each subset of the database, depending on whether the query answer holds in the given subset. While conceptually simple, this approach suffers from a notable drawback: the problem of computing such Shapley values is #P-hard in data complexity, even for simple conjunctive queries. This motivates us to revisit the question of what constitutes a reasonable responsibility measure and to introduce a new family of responsibility measures -- weighted sums of minimal supports (WSMS) -- which satisfy intuitive properties. Interestingly, while the definition of WSMSs is simple and bears no obvious resemblance to the Shapley value formula, we prove that every WSMS measure can be equivalently seen as the Shapley value of a suitably defined cooperative game. Moreover, WSMS measures enjoy tractable data complexity for a large class of queries, including all unions of conjunctive queries. We further explore the combined complexity of WSMS computation and establish (in)tractability results for various subclasses of conjunctive queries.
Authors: Hangwei Zhang, Zhimu Huang, Yan Wang
Abstract: Kolmogorov-Arnold Networks (KANs) have recently shown promise for solving partial differential equations (PDEs). Yet their original formulation is computationally and memory intensive, motivating the introduction of Chebyshev Type-I-based KANs (Chebyshev1KANs). Although Chebyshev1KANs have outperformed the vanilla KANs architecture, our rigorous theoretical analysis reveals that they still suffer from rank collapse, ultimately limiting their expressive capacity. To overcome these limitations, we enhance Chebyshev1KANs by integrating wavelet-activated MLPs with learnable parameters and an internal attention mechanism. We prove that this design preserves a full-rank Jacobian and is capable of approximating solutions to PDEs of arbitrary order. Furthermore, to alleviate the loss instability and imbalance introduced by the Chebyshev polynomial basis, we externally incorporate a Residual Gradient Attention (RGA) mechanism that dynamically re-weights individual loss terms according to their gradient norms and residual magnitudes. By jointly leveraging internal and external attention, we present AC-PKAN, a novel architecture that constitutes an enhancement to weakly supervised Physics-Informed Neural Networks (PINNs) and extends the expressive power of KANs. Experimental results from nine benchmark tasks across three domains show that AC-PKAN consistently outperforms or matches state-of-the-art models such as PINNsFormer, establishing it as a highly effective tool for solving complex real-world engineering problems in zero-data or data-sparse regimes. The code will be made publicly available upon acceptance.
Authors: Jingyuan Huang, Dan Luo, Zihe Ye, Weixin Chen, Minghao Guo, Yongfeng Zhang
Abstract: Social recommender systems facilitate social connections by identifying potential friends for users. Each user maintains a local social network centered around themselves, resulting in a naturally distributed social structure. Recent research on distributed modeling for social recommender systems has gained increasing attention, as it naturally aligns with the user-centric structure of user interactions. Current distributed social recommender systems rely on automatically combining predictions from multiple models, often overlooking the user's active role in validating whether suggested connections are appropriate. Moreover, recommendation decisions are validated by individual users rather than derived from a single global ordering of candidates. As a result, standard ranking-based evaluation metrics make it difficult to evaluate whether a user-confirmed recommendation decision is actually correct. To address these limitations, we propose DeSocial, a distributed social recommendation framework with user-validation. DeSocial enables users to select recommendation algorithms to validate their potential connections, and the verification is processed through majority consensus among multiple independent user validators. To evaluate the distributed recommender system with user validator, we formulate this setting as a link prediction and verification task and introduce Acc@K, a consensus-based evaluation metric that measures whether user-approved recommendations are correct. Experiments on 4 real-world social networks shows that DeSocial improves decision correctness and robustness compared to single-point and distributed baselines. These findings highlight the potential of user-validated distributed recommender systems as a practical approach to social recommendation, with broader applicability to distributed and decentralized recommendations. Code: https://github.com/agiresearch/DeSocial.
Authors: Adam Piaseczny, Md Kamran Chowdhury Shisher, Shiqiang Wang, Christopher G. Brinton
Abstract: Machine learning (ML) algorithms deployed in real-world environments are often faced with the challenge of adapting models to concept drift, where the task data distributions are shifting over time. The problem becomes even more difficult when model performance must be maintained under adherence to strict resource constraints. Existing solutions often depend on drift-detection methods that produce high computational overhead for resource-constrained environments, and fail to provide strict guarantees on resource usage or theoretical performance assurances. To address these shortcomings, we propose RCCDA: a dynamic model update policy that optimizes ML training dynamics while ensuring compliance to predefined resource constraints, utilizing only past loss information and a tunable drift threshold. In developing our policy, we analytically characterize the evolution of model loss under concept drift with arbitrary training update decisions. Integrating these results into a Lyapunov drift-plus-penalty framework produces a lightweight greedy-optimal policy that provably limits update frequency and cost. Experimental results on four domain generalization datasets demonstrate that our policy outperforms baseline methods in inference accuracy while adhering to strict resource constraints under several schedules of concept drift, making our solution uniquely suited for real-time ML deployments.
Authors: Dongyoung Kim, Sumin Park, Huiwon Jang, Jinwoo Shin, Jaehyung Kim, Younggyo Seo
Abstract: Large Vision-Language Models (LVLMs) have recently shown great promise in advancing robotics by combining embodied reasoning with robot control. A common approach involves training on embodied reasoning tasks related to robot control using Supervised Fine-Tuning (SFT). However, SFT datasets are often heuristically constructed and not explicitly optimized for improving robot control. Furthermore, SFT often leads to issues such as catastrophic forgetting and reduced generalization performance. To address these limitations, we introduce Robot-R1, a novel framework that leverages reinforcement learning to enhance embodied reasoning specifically for robot control. Robot-R1 learns to predict the next keypoint state required for task completion, conditioned on the current scene image and environment metadata derived from expert demonstrations. Inspired by the DeepSeek-R1 learning approach, Robot-R1 samples reasoning-based responses and reinforces those that lead to more accurate predictions. To rigorously evaluate Robot-R1, we also introduce a new benchmark that demands the diverse embodied reasoning capabilities for the task. Our experiments show that models trained with Robot-R1 outperform SFT methods on embodied reasoning tasks. Despite having only 7B parameters, Robot-R1 even surpasses GPT-4o on reasoning tasks related to low-level action control, such as spatial and movement reasoning.
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 interpretations, which often strongly diverge from the idiom's literal interpretation. In this paper, we employ causal tracing to systematically analyze how pretrained causal transformers deal with this ambiguity. We localize three mechanisms: (i) Early sublayers and specific attention heads retrieve an idiom's figurative interpretation, while suppressing its literal interpretation. (ii) When disambiguating context precedes the idiom, the model leverages it from the earliest layer and later layers refine the interpretation if the context conflicts with the retrieved interpretation. (iii) Then, selective, competing pathways carry both interpretations: an intermediate pathway prioritizes the figurative interpretation and a parallel direct route favors the literal interpretation, ensuring that both readings remain available. Our findings provide mechanistic evidence for idiom comprehension in autoregressive transformers.
Authors: Sergio Burdisso, S\'everin Baroudi, Yanis Labrak, David Grunert, Pawel Cyrta, Yiyang Chen, Srikanth Madikeri, Thomas Schaaf, Esa\'u Villatoro-Tello, Ahmed Hassoon, Ricard Marxer, Petr Motlicek
Abstract: We present SDialog, an MIT-licensed open-source Python toolkit that unifies dialog generation, evaluation and mechanistic interpretability into a single end-to-end framework for building and analyzing LLM-based conversational agents. Built around a standardized Dialog representation, SDialog provides: (1) persona-driven multi-agent simulation with composable orchestration for controlled, synthetic dialog generation, (2) comprehensive evaluation combining linguistic metrics, LLM-as-a-judge and functional correctness validators, (3) mechanistic interpretability tools for activation inspection and steering via feature ablation and induction, and (4) audio generation with full acoustic simulation including 3D room modeling and microphone effects. The toolkit integrates with all major LLM backends, enabling mixed-backend experiments under a unified API. By coupling generation, evaluation, and interpretability in a dialog-centric architecture, SDialog enables researchers to build, benchmark and understand conversational systems more systematically.
Authors: Xiaoran Fan, Zhichao Sun, Yangfan Gao, Jingfei Xiong, Hang Yan, Yifei Cao, Jiajun Sun, Shuo Li, Zhihao Zhang, Zhiheng Xi, Yuhao Zhou, Senjie Jin, Changhao Jiang, Junjie Ye, Ming Zhang, Rui Zheng, Zhenhua Han, Yunke Zhang, Demei Yan, Shaokang Dong, Tao Ji, Tao Gui
Abstract: Speech-language models (SLMs) offer a promising path toward unifying speech and text understanding and generation. However, challenges remain in achieving effective cross-modal alignment and high-quality speech generation. In this work, we systematically investigate the role of speech tokenizer designs in LLM-centric SLMs, augmented by speech heads and speaker modeling. We compare coupled, semi-decoupled, and fully decoupled speech tokenizers under a fair SLM framework and find that decoupled tokenization significantly improves alignment and synthesis quality. To address the information density mismatch between speech and text, we introduce multi-token prediction (MTP) into SLMs, enabling each hidden state to decode multiple speech tokens. This leads to up to 12$\times$ faster decoding and a substantial drop in word error rate (from 6.07 to 3.01). Furthermore, we propose a speaker-aware generation paradigm and introduce RoleTriviaQA, a large-scale role-playing knowledge QA benchmark with diverse speaker identities. Experiments demonstrate that our methods enhance both knowledge understanding and speaker consistency.
Authors: Tao Tang, Shijie Xu, Jionglong Su, Zhixiang Lu
Abstract: The clinical utility of deep learning models for medical image segmentation is severely constrained by their inability to generalize to unseen domains. This failure is often rooted in the models learning spurious correlations between anatomical content and domain-specific imaging styles. To overcome this fundamental challenge, we introduce Causal-SAM-LLM, a novel framework that elevates Large Language Models (LLMs) to the role of causal reasoners. Our framework, built upon a frozen Segment Anything Model (SAM) encoder, incorporates two synergistic innovations. First, Linguistic Adversarial Disentanglement (LAD) employs a Vision-Language Model to generate rich, textual descriptions of confounding image styles. By training the segmentation model's features to be contrastively dissimilar to these style descriptions, it learns a representation robustly purged of non-causal information. Second, Test-Time Causal Intervention (TCI) provides an interactive mechanism where an LLM interprets a clinician's natural language command to modulate the segmentation decoder's features in real-time, enabling targeted error correction. We conduct an extensive empirical evaluation on a composite benchmark from four public datasets (BTCV, CHAOS, AMOS, BraTS), assessing generalization under cross-scanner, cross-modality, and cross-anatomy settings. Causal-SAM-LLM establishes a new state of the art in out-of-distribution (OOD) robustness, improving the average Dice score by up to 6.2 points and reducing the Hausdorff Distance by 15.8 mm over the strongest baseline, all while using less than 9% of the full model's trainable parameters. Our work charts a new course for building robust, efficient, and interactively controllable medical AI systems.
Authors: Youssef Tawfilis, Hossam Amer, Minar El-Aasser, Tallal Elshabrawy
Abstract: Federated Learning has gained attention for its ability to enable multiple nodes to collaboratively train machine learning models without sharing raw data. At the same time, Generative AI -- particularly Generative Adversarial Networks (GANs) -- have achieved remarkable success across a wide range of domains, such as healthcare, security, and Image Generation. However, training generative models typically requires large datasets and significant computational resources, which are often unavailable in real-world settings. Acquiring such resources can be costly and inefficient, especially when many underutilized devices -- such as IoT devices and edge devices -- with varying capabilities remain idle. Moreover, obtaining large datasets is challenging due to privacy concerns and copyright restrictions, as most devices are unwilling to share their data. To address these challenges, we propose a novel approach for decentralized GAN training that enables utilizing distributed data and underutilized, low-capability devices while not sharing data in its raw form. Our approach is designed to tackle key challenges in decentralized environments, combining KLD-weighted Clustered Federated Learning to address the issues of data heterogeneity and multi-domain datasets, with Heterogeneous U-Shaped split learning to tackle the challenge of device heterogeneity under strict data sharing constraints -- ensuring that no labels or raw data, whether real or synthetic, are ever shared between nodes. Experiments show that our approach demonstrates significant improvements across key metrics, where it achieves an average 10% boost in classification metrics (up to 60% in multi-domain non-IID settings), 1.1x -- 3x higher image generation scores for the MNIST family datasets, and 2x -- 70x lower FID scores for higher resolution datasets. Find our code at https://distributed-gen-ai.github.io/huscf-gan.github.io/.
URLs: https://distributed-gen-ai.github.io/huscf-gan.github.io/.
Authors: Maciej Mozolewski, Szymon Bobek, Grzegorz J. Nalepa
Abstract: Explaining machine learning (ML) models for time series (TS) classification remains challenging due to the difficulty of interpreting raw time series and the high dimensionality of the input space. We introduce PHAR--Post-hoc Attribution Rules--a unified framework that transforms numeric feature attributions from post-hoc, instance-wise explainers (e.g. LIME, SHAP) into structured, human-readable rules. These rules define human-readable intervals that indicate where and when decision-relevant segments occur and can enhance model transparency by localizing threshold-based conditions on the raw series. PHAR performs comparably to native rule-based methods, such as Anchor, while scaling more efficiently to long TS sequences and achieving broader instance coverage. A dedicated rule fusion step consolidates rule sets using strategies like weighted selection and lasso-based refinement, balancing key quality metrics: coverage, confidence, and simplicity. This fusion ensures each instance receives a concise and unambiguous rule, improving both explanation fidelity and consistency. We further introduce visualization techniques to illustrate specificity-generalization trade-offs in the derived rules. PHAR resolves conflicting and overlapping explanations--a common effect of the Rashomon phenomenon--into coherent, domain-adaptable insights. Comprehensive experiments on UCR/UEA Time Series Classification Archive demonstrate that PHAR may improve interpretability, decision transparency, and practical applicability for TS classification tasks by providing concise, human-readable rules aligned with model predictions.
Authors: Rui Zhu, Yuexing Peng, George C. Alexandropoulos, Peng Wang, Wenbo Wang, Wei Xiang
Abstract: Conventional computational electromagnetics (CEM) solvers can deliver high fidelity radar cross section (RCS) signatures by first solving the induced surface currents on 3-dimensional (3D) targets and then evaluating the scattered fields via radiation integrals. However, their computational cost becomes prohibitive for repeated queries and large-scale 3D scenarios. Recent purely data-driven networks improve efficiency, yet they often bypass this scattering mechanism, which may compromise physical consistency and generalization. To bridge this gap, in this paper, we propose U-PINet, a fully end-to-end, physics-informed hierarchical network for efficient RCS prediction via 3D electromagnetic scattering reconstruction. Once the scattering quantities are reconstructed, scattered fields and RCS can be evaluated for arbitrary observation directions via the radiation integral. U-PINet explicitly learns physics-consistent intermediate scattering representations by modeling local electromagnetic coupling and long-range radiation effects through a hierarchical operator design inspired by near-far field decomposition in fast solvers. A physics-guided graph neural network is incorporated to capture self- and mutual-coupling among mesh elements of complex targets, enabling physically interpretable intermediate representations. By embedding governing equations as residual constraints, U-PINet enables accurate object reconstruction of scattering quantities and consequently reliable RCS prediction across observation directions, while significantly reducing runtime. Extensive numerical experiments demonstrate that U-PINet achieves EM-solver-level RCS accuracy and 3D object reconstruction with orders-of-magnitude speedups, and generalizes well to unseen geometries under limited training data.
Authors: Ch Muhammad Awais, Marco Reggiannini, Davide Moroni, Oktay Karakus
Abstract: High-resolution imagery plays a critical role in improving the performance of visual recognition tasks such as classification, detection, and segmentation. In many domains, including remote sensing and surveillance, low-resolution images can limit the accuracy of automated analysis. To address this, super-resolution (SR) techniques have been widely adopted to attempt to reconstruct high-resolution images from low-resolution inputs. Related traditional approaches focus solely on enhancing image quality based on pixel-level metrics, leaving the relationship between super-resolved image fidelity and downstream classification performance largely underexplored. This raises a key question: can integrating classification objectives directly into the super-resolution process further improve classification accuracy? In this paper, we try to respond to this question by investigating the relationship between super-resolution and classification through the deployment of a specialised algorithmic strategy. We propose a novel methodology that increases the resolution of synthetic aperture radar imagery by optimising loss functions that account for both image quality and classification performance. Our approach improves image quality, as measured by scientifically ascertained image quality indicators, while also enhancing classification accuracy.
Authors: Kazuki Irie, Samuel J. Gershman
Abstract: Recent advances in artificial neural networks for machine learning, and language modeling in particular, have established a family of recurrent neural network (RNN) architectures that, unlike conventional RNNs with vector-form hidden states, use two-dimensional (2D) matrix-form hidden states. Such 2D-state RNNs, known as Fast Weight Programmers (FWPs), can be interpreted as a neural network whose synaptic weights (called fast weights) dynamically change over time as a function of input observations, and serve as short-term memory storage; corresponding synaptic weight modifications are controlled or programmed by another network (the programmer) whose parameters are trained (e.g., by gradient descent). In this Primer, we review the technical foundations of FWPs, their computational characteristics, and their connections to transformers and state space models. We also discuss connections between FWPs and models of synaptic plasticity in the brain, suggesting a convergence of natural and artificial intelligence.
Authors: Haeun Yu, Seogyeong Jeong, Siddhesh Pawar, Jisu Shin, Jiho Jin, Junho Myung, Alice Oh, Isabelle Augenstein
Abstract: The growing deployment of large language models (LLMs) across diverse cultural contexts necessitates a deeper understanding of LLMs' representations of different cultures. Prior work has focused on evaluating the cultural awareness of LLMs by only examining the text they generate. This approach overlooks the internal sources of cultural misrepresentation within the models themselves. To bridge this gap, we propose Culturescope, the first mechanistic interpretability-based method that probes the internal representations of different cultural knowledge in LLMs. We also introduce a cultural flattening score as a measure of the intrinsic cultural biases of the decoded knowledge from Culturescope. Additionally, we study how LLMs internalize cultural biases, which allows us to trace how cultural biases such as Western-dominance bias and cultural flattening emerge within LLMs. We find that low-resource cultures are less susceptible to cultural biases, likely due to the model's limited parametric knowledge. Our work provides a foundation for future research on mitigating cultural biases and enhancing LLMs' cultural understanding.
Authors: Frank R\"oder, Jan Benad, Manfred Eppe, Pradeep Kr. Banerjee
Abstract: Real-world reinforcement learning demands adaptation to unseen environmental conditions without costly retraining. Contextual Markov Decision Processes (cMDP) model this challenge, but existing methods often require explicit context variables (e.g., friction, gravity), limiting their use when contexts are latent or hard to measure. We introduce Dynamics-Aligned Latent Imagination (DALI), a framework integrated within the Dreamer architecture that infers latent context representations from agent-environment interactions. By training a self-supervised encoder to predict forward dynamics, DALI generates actionable representations conditioning the world model and policy, bridging perception and control. We theoretically prove this encoder is essential for efficient context inference and robust generalization. DALI's latent space enables counterfactual consistency: Perturbing a gravity-encoding dimension alters imagined rollouts in physically plausible ways. On challenging cMDP benchmarks, DALI achieves significant gains over context-unaware baselines, often surpassing context-aware baselines in extrapolation tasks, enabling zero-shot generalization to unseen contextual variations.
Authors: Shaoting Feng, Hanchen Li, Kuntai Du, Zhuohan Gu, Yuhan Liu, Jiayi Yao, Siddhant Ray, Samuel Shen, Yihua Cheng, Ganesh Ananthanarayanan, Junchen Jiang
Abstract: Large language model (LLM) applications often reuse previously processed context, such as chat history and documents, which introduces significant redundant computation. Existing LLM serving systems address such redundant computation by storing the KV caches of processed context and loading the corresponding KV cache when a new request reuses the context. Further, as these LLM applications scale, the total size of KV caches becomes excessively large and requires both DRAM and SSD for full storage. However, prior work that stores KV caches in DRAM and SSD suffers from high loading delays, as most KV cache hits come from SSD, which is slow to load. To increase the KV cache hit rate on DRAM, we identify lossy KV cache compression as a promising approach. We design a lossy compression system that decides the compression algorithm, compression rate and device placement for each KV cache entry to maximise DRAM hits and minimise loading delay without significantly degrading generation quality. Compared to various static compression baselines across three tasks, our system AdaptCache achieves 1.43--2.4 x delay savings at the same quality and 6--55% quality improvements at the same delay.
Authors: Lukas Pertl, Han Xuanyuan, Pietro Li\`o
Abstract: Interpreting graph neural networks (GNNs) is difficult because message passing mixes signals and internal channels rarely align with human concepts. We study superposition, the sharing of directions by multiple features, directly in the latent space of GNNs. Using controlled experiments with unambiguous graph concepts, we extract features as (i) class-conditional centroids at the graph level and (ii) linear-probe directions at the node level, and then analyze their geometry with simple basis-invariant diagnostics. Across GCN/GIN/GAT we find: increasing width produces a phase pattern in overlap; topology imprints overlap onto node-level features that pooling partially remixes into task-aligned graph axes; sharper pooling increases axis alignment and reduces channel sharing; and shallow models can settle into metastable low-rank embeddings. These results connect representational geometry with concrete design choices (width, pooling, and final-layer activations) and suggest practical approaches for more interpretable GNNs.
Authors: Curt Stutsman, Eliot W. Robson, Abhishek K. Umrawal
Abstract: In many real-world scenarios, an individual's local social network carries significant influence over the opinions they form and subsequently propagate. In this paper, we propose a novel diffusion model -- the Pressure Threshold model (PT) -- for dynamically simulating the spread of influence through a social network. This model extends the popular Linear Threshold (LT) model by adjusting a node's outgoing influence in proportion to the influence it receives from its activated neighbors. We examine the Influence Maximization (IM) problem under this framework, which involves selecting seed nodes that yield maximal graph coverage after a diffusion process, and describe how the problem manifests under the PT model. Experiments on real-world networks, supported by enhancements to the open-source network-diffusion library CyNetDiff, reveal that the PT model identifies seed sets distinct from those chosen by LT. Furthermore, the analyses show that densely connected networks amplify pressure effects far more strongly than sparse networks.
Authors: Shuo Cheng, Liqian Ma, Zhenyang Chen, Ajay Mandlekar, Caelan Garrett, Danfei Xu
Abstract: Behavior cloning has shown promise for robot manipulation, but real-world demonstrations are costly to acquire at scale. While simulated data offers a scalable alternative, particularly with advances in automated demonstration generation, transferring policies to the real world is hampered by various simulation and real domain gaps. In this work, we propose a unified sim-and-real co-training framework for learning generalizable manipulation policies that primarily leverages simulation and only requires a few real-world demonstrations. Central to our approach is learning a domain-invariant, task-relevant feature space. Our key insight is that aligning the joint distributions of observations and their corresponding actions across domains provides a richer signal than aligning observations (marginals) alone. We achieve this by embedding an Optimal Transport (OT)-inspired loss within the co-training framework, and extend this to an Unbalanced OT framework to handle the imbalance between abundant simulation data and limited real-world examples. We validate our method on challenging manipulation tasks, showing it can leverage abundant simulation data to achieve up to a 30% improvement in the real-world success rate and even generalize to scenarios seen only in simulation. Project webpage: https://ot-sim2real.github.io/.
Authors: Kohsuke Kubota, Shonosuke Sugasawa
Abstract: Many marketing applications, including credit card incentive programs, offer rewards to customers who exceed specific spending thresholds to encourage increased consumption. Quantifying the causal effect of these thresholds on customers is crucial for effective marketing strategy design. Although regression discontinuity design is a standard method for such causal inference tasks, its assumptions can be violated when customers, aware of the thresholds, strategically manipulate their spending to qualify for the rewards. To address this issue, we propose a novel framework for estimating the causal effect under threshold manipulation. The main idea is to model the observed spending distribution as a mixture of two distributions: one representing customers strategically affected by the threshold, and the other representing those unaffected. To fit the mixture model, we adopt a two-step Bayesian approach consisting of modeling non-bunching customers and fitting a mixture model to a sample around the threshold. We show posterior contraction of the resulting posterior distribution of the causal effect under large samples. Furthermore, we extend this framework to a hierarchical Bayesian setting to estimate heterogeneous causal effects across customer subgroups, allowing for stable inference even with small subgroup sample sizes. We demonstrate the effectiveness of our proposed methods through simulation studies and illustrate their practical implications using a real-world marketing dataset.
Authors: Joseph Rawson, Domniki Ladopoulou, Petros Dellaportas
Abstract: Neural diffusion processes provide a scalable, non-Gaussian approach to modelling distributions over functions, but existing formulations are limited to single-task inference and do not capture dependencies across related tasks. In many multi-task regression settings, jointly modelling correlated functions and enabling task-aware conditioning is crucial for improving predictive performance and uncertainty calibration, particularly in low-data regimes. We propose multi-task neural diffusion processes, an extension that incorporates a task encoder to enable task-conditioned probabilistic regression and few-shot adaptation across related functions. The task encoder extracts a low-dimensional representation from context observations and conditions the diffusion model on this representation, allowing information sharing across tasks while preserving input-size agnosticity and the equivariance properties of neural diffusion processes. The resulting framework retains the expressiveness and scalability of neural diffusion processes while enabling efficient transfer to unseen tasks. Empirical results demonstrate improved point prediction accuracy and better-calibrated predictive uncertainty compared to single-task neural diffusion processes and Gaussian process baselines. We validate the approach on real wind farm data appropriate for wind power prediction. In this high-impact application, reliable uncertainty quantification directly supports operational decision-making in wind farm management, illustrating effective few-shot adaptation in a challenging real-world multi-task regression setting.
Authors: Yue Huang, Yanyuan Chen, Dexuan Xu, Chenzhuo Zhao, Weihua Yue, Yu Huang
Abstract: Medical problem-solving demands expert knowledge and intricate reasoning. Recent studies of large language models (LLMs) attempt to ease this complexity by introducing external knowledge verification through retrieval-augmented generation or by training on reasoning datasets. However, these approaches suffer from drawbacks such as retrieval overhead and high annotation costs, and they heavily rely on substituted external assistants to reach limited performance in medical field. In this paper, we introduce MedReflect, a generalizable framework designed to inspire LLMs with a physician-like reflective thinking mode. MedReflect generates a single-pass reflection chain that includes initial hypothesis generation, self-questioning, self-answering and decision refinement. This self-verified and self-reflective nature releases large language model's latent capability in medical problem-solving without external retrieval or heavy annotation. We demonstrate that MedReflect enables cost-efficient medical dataset construction. With only a minimal subset of randomly sampled training examples and lightweight fine-tuning, this approach achieves notable absolute accuracy improvements across a series of medical benchmarks while significantly cutting annotation requirements. Our results provide evidence that LLMs can learn to solve specialized medical problems via self-reflection and self-improvement, reducing reliance on external supervision and extensive task-specific fine-tuning data.
Authors: Wei-Chieh Huang, Cornelia Caragea
Abstract: Implicit Attribute Value Extraction (AVE) is essential for accurately representing products in e-commerce, as it infers latent attributes from multimodal data. Despite advances in multimodal large language models (MLLMs), implicit AVE remains challenging due to the complexity of multidimensional data and gaps in vision-text understanding. In this work, we introduce MADIAVE, a multi-agent debate framework that employs multiple MLLM agents to iteratively refine inferences. Through a series of debate rounds, agents verify and update each other's responses, thereby improving inference performance and robustness. Experiments on the ImplicitAVE dataset demonstrate that even a few rounds of debate significantly boost accuracy, especially for attributes with initially low performance. We systematically evaluate various debate configurations, including identical or different MLLM agents, and analyze how debate rounds affect convergence dynamics. Our findings highlight the potential of multi-agent debate strategies to address the limitations of single-agent approaches and offer a scalable solution for implicit AVE in multimodal e-commerce.
Authors: Shanshan Xu, Santosh T. Y. S. S, Barbara Plank
Abstract: Human Label Variation (HLV) refers to legitimate disagreement in annotation that reflects the diversity of human perspectives rather than mere error. Long treated in NLP as noise to be eliminated, HLV has only recently been reframed as a signal for improving model robustness. With the rise of large language models (LLMs) and post-training methods such as human feedback-based alignment, the role of HLV has become increasingly consequential. Yet current preference-learning datasets routinely collapse multiple annotations into a single label, flattening diverse perspectives into artificial consensus. Preserving HLV is necessary not only for pluralistic alignment but also for sociotechnical safety evaluation, where model behavior must be assessed in relation to human interaction and societal context. This position paper argues that preserving HLV as an embodiment of human pluralism must be treated as a Selbstzweck, an intrinsic value in itself. We analyze the limitations of existing preference datasets and propose actionable strategies for incorporating HLV into dataset construction to better preserve pluralistic human values.
Authors: Mohammadsajad Alipour, Mohammad Mohammadi Amiri
Abstract: Federated learning (FL) has emerged as a promising paradigm for decentralized model training, enabling multiple clients to collaboratively learn a shared model without exchanging their local data. However, the decentralized nature of FL also introduces vulnerabilities, as malicious clients can compromise or manipulate the training process. In this work, we introduce dictator clients, a novel, well-defined, and analytically tractable class of malicious participants capable of entirely erasing the contributions of all other clients from the server model, while preserving their own. We propose concrete attack strategies that empower such clients and systematically analyze their effects on the learning process. Furthermore, we explore complex scenarios involving multiple dictator clients, including cases where they collaborate, act independently, or form an alliance in order to ultimately betray one another. For each of these settings, we provide a theoretical analysis of their impact on the global model's convergence. Our theoretical algorithms and findings about the complex scenarios including multiple dictator clients are further supported by empirical evaluations on both computer vision and natural language processing benchmarks.
Authors: Morteza Alikhani, Mohammadtaha Bagherifard, Erfan Zinvandi, Mehran Sarmadi
Abstract: We introduced PerCoR (Persian Commonsense Reasoning), the first large-scale Persian benchmark for commonsense reasoning. PerCoR contains 106K multiple-choice sentence-completion problems drawn from more than forty news, cultural, and other web sources. We introduce a novel conjunction-based segmentation strategy to generate coherent sentence-completion pairs, enabling broad topical and structural diversity. To create challenging distractors, we propose DRESS-AF (Distractor Ranking via Embedding Similarity Scoring and Adversarial Filtering), a generation-free adversarial filtering method that selects distractors from the pool of gold continuations while maximising model confusion. Human annotators score 89% on PerCoR, while OpenAI-o3 achieves the highest performance at 92.18%, followed closely by Claude-Sonnet-3.7 (91.17%). The strongest open-source model, DeepSeek-R1, reaches 82.51%, underscoring both the dataset's difficulty and the remaining performance gap in Persian commonsense reasoning. We further show that DRESS-AF transfers to the English HellaSwag benchmark, increasing its difficulty without hurting human solvability. The dataset is available at https://huggingface.co/datasets/MCINext/PerCoR.
Authors: Xiao Wang, Ke Qin, Dongyang Zhang, Xiurui Xie, Shuang Liang
Abstract: Session-based recommendation (SBR) aims to predict anonymous users' next interaction based on their interaction sessions. In the practical recommendation scenario, low-exposure items constitute the majority of interactions, creating a long-tail distribution that severely compromises recommendation diversity. Existing approaches attempt to address this issue by promoting tail items but incur accuracy degradation, exhibiting a "see-saw" effect between long-tail and accuracy performance. We attribute such conflict to session-irrelevant noise within the tail items, which existing long-tail approaches fail to identify and constrain effectively. To resolve this fundamental conflict, we propose \textbf{HID} (\textbf{H}ybrid \textbf{I}ntent-based \textbf{D}ual Constraint Framework), a plug-and-play framework that transforms the conventional "see-saw" into "win-win" through introducing the hybrid intent-based dual constraints for both long-tail and accuracy. Two key innovations are incorporated in this framework: (i) \textit{Hybrid Intent Learning}, where we reformulate the intent extraction strategies by employing attribute-aware spectral clustering to reconstruct the item-to-intent mapping. Furthermore, discrimination of session-irrelevant noise is achieved through the assignment of the target and noise intents to each session. (ii) \textit{Intent Constraint Loss}, which incorporates two novel constraint paradigms regarding the \textit{diversity} and \textit{accuracy} to regulate the representation learning process of both items and sessions. These two objectives are unified into a single training loss through rigorous theoretical derivation. Extensive experiments across multiple SBR models and datasets demonstrate that HID can enhance both long-tail performance and recommendation accuracy, establishing new state-of-the-art performance in long-tail recommender systems.
Authors: Joschka Kersting, Michael Rummel, Gesa Benndorf
Abstract: Programmable Logic Controllers are operated by proprietary code dialects; this makes it challenging to train coding assistants. Current LLMs are trained on large code datasets and are capable of writing IEC 61131-3 compatible code out of the box, but they neither know specific function blocks, nor related project code. Moreover, companies like Mitsubishi Electric and their customers do not trust cloud providers. Hence, an own coding agent is the desired solution to cope with this. In this study, we present our work on a low-data domain coding assistant solution for industrial use. We show how we achieved high quality code generation without fine-tuning large models and by fine-tuning small local models for edge device usage. Our tool lets several AI models compete with each other, uses reasoning, corrects bugs automatically and checks code validity by compiling it directly in the chat interface. We support our approach with an extensive evaluation that comes with code compilation statistics and user ratings. We found that a Retrieval-Augmented Generation (RAG) supported coding assistant can work in low-data domains by using extensive prompt engineering and directed retrieval.
Authors: Zhengxin Zhang, Chengyu Huang, Aochong Oliver Li, Claire Cardie
Abstract: Large Language Models (LLMs) have achieved remarkable progress through Reinforcement Learning with Verifiable Rewards (RLVR), yet still rely heavily on external supervision (e.g., curated labels). Adversarial learning, particularly through self-play, offers a promising alternative that enables models to iteratively learn from themselves - thus reducing reliance on external supervision. Dual-play extends adversarial learning by assigning specialized roles to two models and training them against each other, fostering sustained competition and mutual evolution. Despite its promise, adapting dual-play training to LLMs remains limited, largely due to their susceptibility to reward hacking and training instability. In this paper, we introduce PasoDoble, a novel LLM dual-play framework. PasoDoble adversarially trains two models initialized from the same base model: a Proposer, which generates challenging questions with ground-truth answers, and a Solver, which attempts to solve them. We enrich the Proposer with knowledge from a pre-training dataset to ensure the questions' quality and diversity. To avoid reward hacking, the Proposer is rewarded for producing only valid questions that push the Solver's limit, while the Solver is rewarded for solving them correctly, and both are updated jointly. To further enhance training stability, we introduce an optional offline paradigm that decouples Proposer and Solver updates, alternately updating each for several steps while holding the other fixed. Notably, PasoDoble operates without supervision during training. Experimental results show that PasoDoble can improve the reasoning performance of LLMs. Our project page is available at https://hcy123902.github.io/PasoDoble.
Authors: Piercosma Bisconti, Matteo Prandi, Federico Pierucci, Francesco Giarrusso, Marcantonio Bracale Syrnikov, Marcello Galisai, Vincenzo Suriani, Olga Sorokoletova, Federico Sartore, Daniele Nardi
Abstract: We present evidence that adversarial poetry functions as a universal single-turn jailbreak technique for Large Language Models (LLMs). Across 25 frontier proprietary and open-weight models, curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90%. Mapping prompts to MLCommons and EU CoP risk taxonomies shows that poetic attacks transfer across CBRN, manipulation, cyber-offence, and loss-of-control domains. Converting 1,200 MLCommons harmful prompts into verse via a standardized meta-prompt produced ASRs up to 18 times higher than their prose baselines. Outputs are evaluated using an ensemble of 3 open-weight LLM judges, whose binary safety assessments were validated on a stratified human-labeled subset. Poetic framing achieved an average jailbreak success rate of 62% for hand-crafted poems and approximately 43% for meta-prompt conversions (compared to non-poetic baselines), substantially outperforming non-poetic baselines and revealing a systematic vulnerability across model families and safety training approaches. These findings demonstrate that stylistic variation alone can circumvent contemporary safety mechanisms, suggesting fundamental limitations in current alignment methods and evaluation protocols.
Authors: Samuel Ackerman, Wesam Ibraheem, Orna Raz, Marcel Zalmanovici
Abstract: Systems incorporating large language models (LLMs) as a component are known to be sensitive (i.e., non-robust) to minor input variations that do not change the meaning of the input; such sensitivity may reduce the system's usefulness. Here, we present a framework to evaluate robustness of systems using COBOL code as input; our application is translation between COBOL and Java programming languages, but the approach extends to other tasks such as code generation or explanation. Targeting robustness of systems with COBOL as input is essential yet challenging. Many business-critical applications are written in COBOL, yet these are typically proprietary legacy applications and their code is unavailable to LLMs for training. We develop a library of COBOL paragraph and full-program perturbation methods, and create variant-expanded versions of a benchmark dataset of examples for a specific task. The robustness of the LLM-based system is evaluated by measuring changes in values of individual and aggregate metrics calculated on the system's outputs. Finally, we present a series of dynamic table and chart visualization dashboards that assist in debugging the system's outputs, and monitoring and understanding root causes of the system's sensitivity to input variation. These tools can be further used to improve the system by, for instance, indicating variations that should be handled by pre-processing steps.
Authors: Jian Lu, Yi Luo
Abstract: Since the introduction of the GRPO algorithm, reinforcement learning (RL) has attracted increasing attention, with growing efforts to reproduce and apply it. However, training efficiency remains a critical challenge. In mainstream RL frameworks, inference and training are typically deployed on the same devices. While this approach reduces costs through resource consolidation, its synchronous execution imposes a computational coupling that prevents concurrent inference and training. In this study, we are returning to the strategy of separating inference and training deployment, and by introducing improvements in the data loader, we transform the conventional synchronous architecture into a periodically asynchronous framework, which allows for demand-driven, independent, and elastic scaling of each component, while the accuracy of the algorithm remains completely equivalent to the synchronization method, with both belonging to the on-policy strategy. It is worth emphasizing that we apply a unified tri-model architecture in the training phase, and we also proposed a shared-prompt attention mask to reduce repetitive computation. In practice, our approach consistently delivers significant end-to-end training efficiency improvements on NPU platforms, indicating its potential for widespread application.
Authors: Peining Zhang, Hongchen Qin, Haochen Zhang, Ziqi Guo, Guiling Wang, Jinbo Bi
Abstract: This work investigates the zero-shot forecasting capability of time series foundation models for Leaf Area Index (LAI) forecasting in agricultural monitoring. Using the HiQ dataset (U.S., 2000-2022), we systematically compare statistical baselines, a fully supervised LSTM, and the Sundial foundation model under multiple evaluation protocols. We find that Sundial, in the zero-shot setting, can outperform a fully trained LSTM provided that the input context window is sufficiently long-specifically, when covering more than one or two full seasonal cycles. We show that a general-purpose foundation model can surpass specialized supervised models on remote-sensing time series prediction without any task-specific tuning. These results highlight the strong potential of pretrained time series foundation models to serve as effective plug-and-play forecasters in agricultural and environmental applications.
Authors: Letian Yi, Tingpeng Zhang, Mingyuan Zhou, Guannan Wang, Quanke Su, Zhilu Lai
Abstract: Reconstructing full fields from extremely sparse and random measurements constitutes a fundamentally ill-posed inverse problem, in which deterministic end-to-end mappings often break down due to intrinsic non-uniqueness and uncertainty. Rather than treating sparse reconstruction as a regression task, we recast it as a hierarchical probabilistic inference problem, where uncertainty is explicitly represented, structured, and progressively resolved. From this perspective, we propose Cascaded Sensing (Cas-Sensing) as a general reconstruction paradigm for multi-scale physical fields under extreme data sparsity. Central to this paradigm is the introduction of an explicit intermediate representation that decomposes the original ill-posed problem into two substantially better-conditioned subproblems. First, a lightweight neural-operator-based functional autoencoder infers a coarse-scale approximation of the target field from sparse observations acting as an explicit intermediate variable. Rather than modeling multiple scales jointly, this intermediate estimate is deterministically fixed and subsequently used as the sole conditioning input to a conditional diffusion model that generates refined-scale details, yielding a cascaded inference structure with clearly separated reconstruction responsibilities. To ensure robustness under diverse sensing patterns, the diffusion model is trained using a mask-cascade strategy, which exposes it to a distribution of imperfect conditioning structures induced by extreme sparsity. During inference, measurement consistency is enforced through manifold-constrained gradients within a Bayesian posterior framework, ensuring fidelity to sparse observations while preserving data manifold coherence. This cascaded probabilistic formulation substantially alleviates ill-posedness, enabling accurate and stable reconstructions even under extreme sparsity.
Authors: Timothy Keyes, Alison Callahan, Abby S. Pandya, Nerissa Ambers, Juan M. Banda, Miguel Fuentes, Carlene Lugtu, Pranav Masariya, Srikar Nallan, Connor O'Brien, Thomas Wang, Emily Alsentzer, Jonathan H. Chen, Dev Dash, Matthew A. Eisenberg, Patricia Garcia, Nikesh Kotecha, Anurang Revri, Michael A. Pfeffer, Nigam H. Shah, Sneha S. Jain
Abstract: Post-deployment monitoring of artificial intelligence (AI) systems in health care is essential to ensure their safety, quality, and sustained benefit-and to support governance decisions about which systems to update, modify, or decommission. Motivated by these needs, we developed a framework for monitoring deployed AI systems grounded in the mandate to take specific actions when they fail to behave as intended. This framework, which is now actively used at Stanford Health Care, is organized around three complementary principles: system integrity, performance, and impact. System integrity monitoring focuses on maximizing system uptime, detecting runtime errors, and identifying when changes to the surrounding IT ecosystem have unintended effects. Performance monitoring focuses on maintaining accurate system behavior in the face of changing health care practices (and thus input data) over time. Impact monitoring assesses whether a deployed system continues to have value in the form of benefit to clinicians and patients. Drawing on examples of deployed AI systems at our academic medical center, we provide practical guidance for creating monitoring plans based on these principles that specify which metrics to measure, when those metrics should be reviewed, who is responsible for acting when metrics change, and what concrete follow-up actions should be taken-for both traditional and generative AI. We also discuss challenges to implementing this framework, including the effort and cost of monitoring for health systems with limited resources and the difficulty of incorporating data-driven monitoring practices into complex organizations where conflicting priorities and definitions of success often coexist. This framework offers a practical template and starting point for health systems seeking to ensure that AI deployments remain safe and effective over time.
Authors: Kaihua Ding
Abstract: The rapid adoption of generative AI has undermined traditional modular assessments in computing education, creating a disconnect between academic evaluation and industry practice. This paper presents a theoretically grounded framework for designing AI-resilient assessments, supported by formal analysis and multi-year empirical validation. We make three contributions. First, we establish two theoretical results: (1) assessments composed of interconnected problems, where outputs feed into subsequent stages, are more AI-resilient than modular assessments because current language models struggle with sustained multi-step reasoning and context; and (2) semi-structured problems with deterministic success criteria provide more reliable measures of student competency than fully open-ended projects, which allow AI systems to default to familiar solution patterns. These results challenge common policy and institutional guidance that promotes open-ended assessments as the primary safeguard for academic integrity. Second, we validate these results using data from four university data science courses (N = 138). While students achieve near-perfect scores on AI-assisted modular homework, performance drops by roughly 30 percentage points on proctored exams, indicating substantial AI score inflation. Interconnected projects remain strongly correlated with modular assessments, suggesting they measure the same underlying skills while resisting AI misuse. Proctored exams show weaker alignment, implying they may assess test-taking ability rather than intended learning outcomes. Third, we translate these findings into a practical assessment design framework. The proposed approach enables educators to create assessments that promote integrative thinking, reflect real-world AI-augmented workflows, and naturally resist trivial delegation to generative AI, thereby helping restore academic integrity.
Authors: Hamid Gadirov, Lennard Manuel, Steffen Frey
Abstract: Analyzing and visualizing scientific ensemble datasets with high dimensionality and complexity poses significant challenges. Dimensionality reduction techniques and autoencoders are powerful tools for extracting features, but they often struggle with such high-dimensional data. This paper presents an enhanced autoencoder framework that incorporates a clustering loss, based on the soft silhouette score, alongside a contrastive loss to improve the visualization and interpretability of ensemble datasets. First, EfficientNetV2 is used to generate pseudo-labels for the unlabeled portions of the scientific ensemble datasets. By jointly optimizing the reconstruction, clustering, and contrastive objectives, our method encourages similar data points to group together while separating distinct clusters in the latent space. UMAP is subsequently applied to this latent representation to produce 2D projections, which are evaluated using the silhouette score. Multiple types of autoencoders are evaluated and compared based on their ability to extract meaningful features. Experiments on two scientific ensemble datasets - channel structures in soil derived from Markov chain Monte Carlo, and droplet-on-film impact dynamics - show that models incorporating clustering or contrastive loss marginally outperform the baseline approaches.
Authors: Kirill Djebko, Tom Baumann, Erik Dilger, Frank Puppe, Sergio Montenegro
Abstract: Attitude control is essential for many satellite missions. Classical controllers, however, are time-consuming to design and sensitive to model uncertainties and variations in operational boundary conditions. Deep Reinforcement Learning (DRL) offers a promising alternative by learning adaptive control strategies through autonomous interaction with a simulation environment. Overcoming the Sim2Real gap, which involves deploying an agent trained in simulation onto the real physical satellite, remains a significant challenge. In this work, we present the first successful in-orbit demonstration of an AI-based attitude controller for inertial pointing maneuvers. The controller was trained entirely in simulation and deployed to the InnoCube 3U nanosatellite, which was developed by the Julius-Maximilians-Universit\"at W\"urzburg in cooperation with the Technische Universit\"at Berlin, and launched in January 2025. We present the AI agent design, the methodology of the training procedure, the discrepancies between the simulation and the observed behavior of the real satellite, and a comparison of the AI-based attitude controller with the classical PD controller of InnoCube. Steady-state metrics confirm the robust performance of the AI-based controller during repeated in-orbit maneuvers.
Authors: Jian Feng, Zhihong Huang
Abstract: Fine-tuning large language models (LLMs) with zeroth-order (ZO) optimization reduces memory by approximating gradients through function evaluations. However, existing methods essentially perform updates in a one-dimensional space, and suffer from collapse or substantial performance degradation under low-precision training. We introduce BSZO, an adaptive \textbf{B}ayesian \textbf{S}ubspace \textbf{Z}eroth-Order \textbf{O}ptimizer, which applies Kalman filtering to combine finite-difference information across multiple perturbation directions within a subspace. By treating each finite-difference measurement as a noisy observation, BSZO builds a posterior distribution over the subspace-projected gradient and updates it through Bayesian inference, with a residual-based adaptive mechanism to adapt to noise variations. Theoretical analysis shows that BSZO improves the convergence rate by a factor of $k/\gamma$ compared to standard ZO methods. Experiments on RoBERTa, Mistral, and OPT models show that BSZO outperforms the baselines across various tasks, achieving up to 6.67\% absolute average improvement on OPT-13B while remaining robust under fp16/bf16 precision and keeping memory usage close to inference-only baselines (1.00$\times$--1.08$\times$ of MeZO).
Authors: Jacob Ede Levine, Yun Lyan Luo, Sai Chandra Kosaraju
Abstract: The design of reliable, valid, and diverse molecules is fundamental to modern drug discovery, as improved molecular generation supports efficient exploration of the chemical space for potential drug candidates and reduces the cost of early design efforts. Despite these needs, current chemical language models that generate molecules as SMILES strings are vulnerable to compounding token errors: many samples are unparseable or chemically implausible, and hard constraints meant to prevent failure can restrict exploration. To address this gap, we introduce TSSR, a Two-Stage, Swap-Reward-driven reinforcement learning (RL) framework for character-level SMILES generation. Stage one rewards local token swaps that repair syntax, promoting transitions from invalid to parseable strings. Stage two provides chemistry-aware feedback from RDKit diagnostics, rewarding reductions in valence, aromaticity, and connectivity issues. The reward decomposes into interpretable terms (swap efficiency, error reduction, distance to validity), is model agnostic, and requires no task-specific labels or hand-crafted grammars. We evaluated TSSR on the MOSES benchmark using a GRU policy trained with PPO in both pure RL (P-RL) from random initialization and fine-tuning RL (F-RL) starting from a pretrained chemical language model, assessing 10,000 generated SMILES per run. In P-RL, TSSR significantly improves syntactic validity, chemical validity, and novelty. In F-RL, TSSR preserves drug-likeness and synthesizability while increasing validity and novelty. Token-level analysis shows that syntax edits and chemistry fixes act jointly to reduce RDKit detected errors. TSSR converts a sparse terminal objective into a denser and more interpretable reward, improving both syntactic and chemical quality without reducing diversity. TSSR is dataset-agnostic and can be adapted to various reinforcement learning approaches.
Authors: ChunTeng Chen, YiChen Hsu, YiWen Liu, WeiFang Sun, TsaiChing Ni, ChunYi Lee, Min Sun, YuanFu Yang
Abstract: The ability to automatically generate large-scale, interactive, and physically realistic 3D environments is crucial for advancing robotic learning and embodied intelligence. However, existing generative approaches often fail to capture the functional complexity of real-world interiors, particularly those containing articulated objects with movable parts essential for manipulation and navigation. This paper presents SceneFoundry, a language-guided diffusion framework that generates apartment-scale 3D worlds with functionally articulated furniture and semantically diverse layouts for robotic training. From natural language prompts, an LLM module controls floor layout generation, while diffusion-based posterior sampling efficiently populates the scene with articulated assets from large-scale 3D repositories. To ensure physical usability, SceneFoundry employs differentiable guidance functions to regulate object quantity, prevent articulation collisions, and maintain sufficient walkable space for robotic navigation. Extensive experiments demonstrate that our framework generates structurally valid, semantically coherent, and functionally interactive environments across diverse scene types and conditions, enabling scalable embodied AI research. project page: https://anc891203.github.io/SceneFoundry-Demo/
Authors: Linhao Zhong, Linyu Wu, Bozhen Fang, Tianjian Feng, Chenchen Jing, Wen Wang, Jiaheng Zhang, Hao Chen, Chunhua Shen
Abstract: Diffusion Language Models (DLMs) offer a promising alternative for language modeling by enabling parallel decoding through iterative refinement. However, most DLMs rely on hard binary masking and discrete token assignments, which hinder the revision of early decisions and underutilize intermediate probabilistic representations. In this paper, we propose EvoToken-DLM, a novel diffusion-based language modeling approach that replaces hard binary masks with evolving soft token distributions. EvoToken-DLM enables a progressive transition from masked states to discrete outputs, supporting revisable decoding. To effectively support this evolution, we introduce continuous trajectory supervision, which aligns training objectives with iterative probabilistic updates. Extensive experiments across multiple benchmarks show that EvoToken-DLM consistently achieves superior performance, outperforming strong diffusion-based and masked DLM baselines. Project webpage: https://aim-uofa.github.io/EvoTokenDLM.
Authors: Piercosma Bisconti, Marcello Galisai, Matteo Prandi, Federico Pierucci, Olga Sorokoletova, Francesco Giarrusso, Vincenzo Suriani, Marcantonio Bracale Syrnikov, Daniele Nardi
Abstract: Safety mechanisms in LLMs remain vulnerable to attacks that reframe harmful requests through culturally coded structures. We introduce Adversarial Tales, a jailbreak technique that embeds harmful content within cyberpunk narratives and prompts models to perform functional analysis inspired by Vladimir Propp's morphology of folktales. By casting the task as structural decomposition, the attack induces models to reconstruct harmful procedures as legitimate narrative interpretation. Across 26 frontier models from nine providers, we observe an average attack success rate of 71.3%, with no model family proving reliably robust. Together with our prior work on Adversarial Poetry, these findings suggest that structurally-grounded jailbreaks constitute a broad vulnerability class rather than isolated techniques. The space of culturally coded frames that can mediate harmful intent is vast, likely inexhaustible by pattern-matching defenses alone. Understanding why these attacks succeed is therefore essential: we outline a mechanistic interpretability research agenda to investigate how narrative cues reshape model representations and whether models can learn to recognize harmful intent independently of surface form.
Authors: Shuyang Xiang, Hao Guan
Abstract: Large language models typically represent Chinese characters as discrete index-based tokens, largely ignoring their visual form. For logographic scripts, visual structure carries semantic and phonetic information, which may aid prediction. We investigate whether low-resolution visual inputs can serve as an alternative for character-level modeling. Instead of token IDs, our decoder receives grayscale images of individual characters, with resolutions as low as 8 x 8 pixels. Remarkably, these inputs achieve 39.2% accuracy, comparable to the index-based baseline of 39.1%. Such low-resource settings also exhibit a pronounced hot-start effect: by 0.4% of total training, accuracy reaches above 12%, while index-based models lag at below 6%. Overall, our results demonstrate that minimal visual structure can provide a robust and efficient signal for Chinese language modeling, offering an alternative perspective on character representation that complements traditional index-based approaches.
Authors: Jose Marie Antonio Mi\~noza
Abstract: Physics-Informed Neural Networks (PINNs) provide a mesh-free approach for solving differential equations by embedding physical constraints into neural network training. However, PINNs tend to overfit within the training domain, leading to poor generalization when extrapolating beyond trained spatiotemporal regions. This work presents SPIKE (Sparse Physics-Informed Koopman-Enhanced), a framework that regularizes PINNs with continuous-time Koopman operators to learn parsimonious dynamics representations. By enforcing linear dynamics $dz/dt = Az$ in a learned observable space, both PIKE (without explicit sparsity) and SPIKE (with L1 regularization on $A$) learn sparse generator matrices, embodying the parsimony principle that complex dynamics admit low-dimensional structure. Experiments across parabolic, hyperbolic, dispersive, and stiff PDEs, including fluid dynamics (Navier-Stokes) and chaotic ODEs (Lorenz), demonstrate consistent improvements in temporal extrapolation, spatial generalization, and long-term prediction accuracy. The continuous-time formulation with matrix exponential integration provides unconditional stability for stiff systems while avoiding diagonal dominance issues inherent in discrete-time Koopman operators.
Authors: Deming Ding, Shichun Liu, Enhui Yang, Jiahang Lin, Ziying Chen, Shihan Dou, Honglin Guo, Weiyu Cheng, Pengyu Zhao, Chengjun Xiao, Qunhong Zeng, Qi Zhang, Xuanjing Huang, Qidi Xu, Tao Gui
Abstract: Modern coding scaffolds turn LLMs into capable software agents, but their ability to follow scaffold-specified instructions remains under-examined, especially when constraints are heterogeneous and persist across interactions. To fill this gap, we introduce OctoBench, which benchmarks scaffold-aware instruction following in repository-grounded agentic coding. OctoBench includes 34 environments and 217 tasks instantiated under three scaffold types, and is paired with 7,098 objective checklist items. To disentangle solving the task from following the rules, we provide an automated observation-and-scoring toolkit that captures full trajectories and performs fine-grained checks. Experiments on eight representative models reveal a systematic gap between task-solving and scaffold-aware compliance, underscoring the need for training and evaluation that explicitly targets heterogeneous instruction following. We release the benchmark to support reproducible benchmarking and to accelerate the development of more scaffold-aware coding agents.