The 2nd Workshop on

Compositional Learning: Safety, Interpretability, and Agents

ICML 2026, COEX Convention & Exhibition Center (Auditorium), July 11th, 2026

Compositionality, defined as the ability to construct and reason about complex concepts from reusable components, is a hallmark of human cognition and the key to robust generalization. Despite the astonishing progress of modern AI systems, it remains an open question whether they truly capture and leverage the compositional nature of many real-world domains. The workshop explores this pressing challenge across multiple critical dimensions. We invite contributions focusing on the theoretical foundations of compositionality, its central role in the age of foundation models and agents, and its impact on achieving robustness and systematic out-of-domain generalization. Among the most promising topics related to compositionality, we identify three timely and impactful foci for this edition of the workshop.

• How can systems generalize beyond training distributions to improve safety? Exploring the representational structures and learning dynamics that enable true compositional understanding under distribution shifts. This involves examining the roles of inductive biases, abstraction, and modularity to identify strategies that allow systems to generalize in scenarios where simple statistical correlations fail, hence improving safe deployment in real-world scenarios.
• How do models internally represent and learn compositionality? Investigating the mechanisms by which LLMs represent, acquire, and generalize compositional structures. This focuses on interpreting internal states, analyzing emergent behaviors, and establishing rigorous benchmarks to evaluate whether models leverage compositionality or rely on memorization.
• How can agents leverage compositionality for complex tasks? Studying how compositional principles drive the development of robust, generalizable agents. Key areas include the composition of skills and sub-goals for long-horizon planning, the systematic integration of retrieved knowledge and tools, and the creation of architectures that dynamically combine neural and symbolic modules to execute multi-step tasks.

Furthermore, through an interdisciplinary dialogue featuring high-profile guests from both academia and industry, we aim to catalyze new research directions that push the boundaries of compositional learning in advanced AI systems.

Speakers and Panelists

Call for Papers

Authors are welcome to submit a 4-page or 8-page (short/long) paper based on in-progress work, or a relevant paper being presented at the main conference, in the following topics:

• Compositionality in Action: Agentic AI, Planning, and Tool Use
• Safety: architectures and representations for robust and generalizable systems
• Explainability: representations and reasoning in foundation models
• Theoretical foundations and general principles of compositionality in AI
• Multimodal compositionality
• Modular and dynamic architectures
• Continual/transfer learning through compositionality
• Compositional learning for various application domains, such as computer vision, natural language processing, reinforcement learning, and science

We welcome review and positional papers that may foster discussions. Accepted papers will be presented during poster sessions, with exceptional submissions selected for spotlight oral presentations. Accepted papers will be made publicly available as non-archival reports, allowing future submissions to archival conferences or journals. The following dates are important for the submission:

• Paper submissions open: March 30, 2026, AOE
• Paper submissions deadline: May 7, 2026, AOE
• Notification to authors: May 22, 2026, AOE

Submission Guidelines

• Submission URL: https://openreview.net/group?id=ICML.cc/2026/Workshop/CompLearn
• Format: All submissions must be in PDF format and anonymized. Submissions are limited to 4/8 content pages, including all figures and tables; unlimited additional pages containing references and supplementary materials are allowed. Reviewers may choose to read the supplementary materials but will not be required to. Camera-ready versions may go up to 5/9 content pages.
• Style file: You must format your submission using the ICML 2026 LaTeX style file. The maximum file size for submissions is 20MB. Submissions that violate the ICML style (e.g., by decreasing margins or font sizes) or page limits may be rejected without further review.
• Dual-submission policy: We welcome ongoing and unpublished work. We will also accept papers that are under review at the time of submission, or that have been recently accepted without published proceedings.
• Non-archival: The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.
• Visibility: Submissions and reviews will not be public. Only accepted papers will be made public.
• Mandatory Reciprocal Reviewing: At least one author per submission must commit to reviewing for the workshop. You will need to designate the reviewing author(s) on OpenReview. Submissions without a nominated reviewer may be desk-rejected.
• Contact: For any questions, please contact us at compositional-learning-icml-2026@googlegroups.com.

If you would like to become a reviewer for this workshop, please let us know https://forms.gle/nG4idePAF4Qp6TNk8.

Awards

We are pleased to announce that GuideLabs is sponsoring this year's awards. Spotlight nominees will receive a $300 prize, and the Best Reviewer Award is $150. Furthermore, Cambridge University Press is offering a book from the Cambridge catalog of the prize winner's choice up to a value of $100 to every author and reviewer receiving an award.

08:25 AM	Opening Remarks
08:30 AM	Spotlight Talks Doojin Baek, Gyubin Lee, Junyeob Baek, Hosung Lee, Sungjin Ahn. Learning to Theorize the World from Observation. Patrick Knab, Sascha Marton, Philipp Johannes Schubert, Drago Andres Guggiana Nilo, Christian Bartelt. Concepts in Motion: Temporal Concept Bottleneck Model for Interpretable Video Classification. Nikhil Garg, Jon Kleinberg, Kenny Peng. How Many Features Can a Language Model Store Under the Linear Representation Hypothesis?
9:00 AM	Invited Talk: Nouha Dziri (Senior Research Scientist, Cohere) Understanding and Improving Reasoning in Large Reasoning Models.
9:35 AM	Invited Talk: David Cox (VP for AI models, IBM) Generative Computing and the Future of AI Software Development.
10:10 AM	Coffee Break
10:20 AM	Invited Talk: Parisa Kordjamshidi (Associate Professor, Michigan State University) Compositional Reasoning for Natural Language Comprehension and Grounding Leveraging Neuro-Symbolic AI.
10:55 AM	Invited Talk: Aya Abdelsalam Ismail (CTO, Guide Labs) Interpretability by Design: Why We Should Build Models We Can Understand.
11:25 AM	Awards cerimony
11:30 AM	Poster Session 1 [More] Pavan C Shekar, Aswanth Krishnan. Adaptive Minds: Empowering Agents with LoRAs as Tools. #2311 Jerry Yao-Chieh Hu, Hong-Yu Chen, Po-Chiao Lin, Maojiang Su, Han Liu. In-Context Universal Approximation, Compositional Generalization, and Algorithm Emulation. #2311 Augustus Haoyang Li. THEIA: Learning Complete Kleene Three-Valued Logic in a Pure-Neural Modular Architecture. #2312 Hong-Yu Chen, Venkat Sripad Ganti, Hude Liu, Jerry Yao-Chieh Hu, Han Liu. Chain-of-Thought Gradient Descent. #2312 Rundong Yang. From Self-Preservation to Peer-Preservation: A Staged Framing of Preservation-Oriented Misalignment in Frontier Models. #2313 Jin Yoo. Nouns, Not Modifiers: OpenVLA Parses Objects but Fails at Spatial Composition. #2313 Jason Yoo, Rajarshi Saha, Shaowei Zhu, Tao Yu, Wei Tang, Youngsuk Park. MKEvolve: A Modular Multi-Agent Framework for Kernel Code Generation. #2314 Mihyang Kim. Compositional Skill Execution in LLM Multi-Agent Systems: A Comparative Study of Collaboration Architectures for Long-Horizon Tasks. #2314 Anna-Alina Bondarets, Taras Rumezhak, Volodymyr Karpiv. LGPro: Language-Guided Prototype Discovery for Compositional Zero-Shot Learning. #2315 Hong-Jie You, Jie-Jing Shao, Xiao-Wen Yang, Zhi-Fan Wu, Lin-Han Jia, Lan-Zhe Guo, Yu-Feng Li. FormalImG: Evaluating Structural Compositional Generalization for T2I Models. #2315 Ashish Baghel. Compositional Evolutionary Probing of LLM Safety Alignment. #2400 Gael Gendron, Joze M. Rozanec, Michael J. Witbrock, Gillian Dobbie. Causal Cartographer: From Mapping to Reasoning Over Counterfactual Worlds. #2400 Aryan Sharma, Cutter Dawes, Shivam Raval. Dissociating Decodability and Causal Use in Bracket-Sequence Transformers. # Young-Min Kim, Bo-Yeong Kang. Compositional Skill Chaining and Policy Blending for Hard Exploration in the BRIO Labyrinth Game. #2401 Nathaniel Bottman, Kyle Richardson. Operads for compositional reasoning in LLMs. #2402 Hengyu Liu, Chenxin Li, Wenbo Hu, Zhiqin Yang, Yuxin Chen, Ying Shan, Brandon Y. Feng. Struct-to-Reason: Enhancing Video Understanding of Vision-Language Models by Decoupling Perception and Reasoning via Structured Summary. #2402 Luca Miglior, Vincenzo Gervasi, Davide Bacciu. CompFlow: Composing Velocity Fields for Multi-Condition Generation. #2403 Stjepan Požgaj, Dobrik Georgiev Georgiev, Marin Šilić, Goran Delac, Klemo Vladimir. Fixing Edge-Pointer Compositionality in Neural Algorithmic Reasoning. #2403 Kyeong Seon Kim, Jiyeon Son, Tae-Hyun Oh. IGG: A Benchmark for Interactive GUI Grounding under Visibility Constraints. #2404 Manan Gupta, Chinmay Pushkar, Sanchit Kabra, Dhruv Kumar, Jagat Sesh Challa. MultiVulnBench: A Large-Scale Benchmark for Count Bias in LLM-Based Multi-Vulnerability Detection. #2404 Moon Ye-Bin, Jiyeon Son, Tae-Hyun Oh. Spatially Stable GUI Grounding via Zoom Consistency Loss. #2405 Chengxin Liu, Moon Ye-Bin, Tae-Hyun Oh. Entropy-Aware GUI Grounding: From Failure Analysis to Improved Localization. #2405 Patrick Knab, Sascha Marton, Philipp Johannes Schubert, Drago Andres Guggiana Nilo, Christian Bartelt. Concepts in Motion: Temporal Concept Bottleneck Model for Interpretable Video Classification. #2406 Leo Lorence George, Anushri Iyer, Abhishek Bakshi, Pavan Kulkarni. Attractor Inversion: A Geometric Account of Adversarial Manipulation in Human Decision-Making. #2406 Manan Gupta, Dhruv Kumar. DPMI: A Principled Index for Neural Polysemanticity via Dirichlet Process Mixture Modeling. #2407 Manan Gupta, Dhruv Kumar. Reasoning Phases Are Continuous, Not Discrete: Evidence from Switching Linear Dynamical Systems Applied to Chain-of-Thought Residual Streams. #2407 Pietro Barbiero, Mateo Espinosa Zarlenga, Francesco Giannini, Alberto Termine, Filippo Bonchi, Mateja Jamnik, Giuseppe Marra. Actionable Interpretability Must Be Defined in Terms of Symmetries: A Compositional Probabilistic Approach. #2408 Yassine Taoudi-Benchekroun, Klim Troyan, Pascal Josef Sager, Stefan Gerber, Lukas Tuggener, Thilo Stadelmann, Benjamin F Grewe. COGITAO: A Procedural and Object-Centric Framework to Evaluate Compositional and Systematic Generalization. #2408 Zhiwei Shang, Jiahang Sun, Mingrong Gong, Mingze Kong, Zikun Qu, Pingchen Lu, Junhao Dong, Zhipiao Liu, Hongwei Yang, XIE GUOQING, Yao Shu, Zhongxiang Dai. Fusion is the New Mutation: Bandit-Guided Evolution on Workflow Graphs. #2409 Beomgwon Jo, Sunchan Park, Kyeongbo Kong. Compositional Agentic Formulation Search for Open-Vocabulary Audio-Visual Event Localization. #2409 Ward Gauderis, Thomas Dooms, Steven T. Homer, Kola Ayonrinde, Geraint A. Wiggins. From Mechanistic to Compositional Interpretability. #2410 Siddharth Karuturi, Kaustubh S. Bukkapatnam, Soham Batra, Laksh Patel, Tanush Ajay Shastry. The Spurious Composition Problem: Conditional Independence as a Necessary and Sufficient Condition for Systematic Generalization. #2410 Siddharth Karuturi, Kaustubh S. Bukkapatnam, Tanush Ajay Shastry. When Does Composition Compose? A PAC-Theoretic Framework for Compositional Faithfulness, Safety Certificates, and Training Dynamics. #2411 Changho Shin, Daiwei Chen, John Cooper, Brenden Lake, Frederic Sala, Ramya Korlakai Vinayak. Compositional Self-Improvement. #2411 Satwik Bathula, Anand Joshi. Gating Enables Curvature: A Geometric Expressivity Gap in Attention. #2412 Yujin Jeong, Arnas Uselis, Iro Laina, Seong Joon Oh, Anna Rohrbach. When Do Diffusion Models learn to Generate Multiple Objects?. #2412 Kaustubh S. Bukkapatnam, Siddharth Karuturi. Circuit Modularity Predicts Compositional Generalization: Theory and Evidence from Transformers. #2413 Rishi Ashish Shah, Shivaay Dhondiyal, Sarthak Pandey. When Does Disentanglement Enable Compositional Generalization? A Transfer Bound and Its Empirical Validation. #2413 Junwei Su, Chuan Wu. When Do Multi-Agent Systems Outperform? Analysing the Learning Efficiency of Agentic Systems. #2414 Yue Huang, Yu Jiang, Wenjie Wang, Haomin Zhuang, Xiaonan Luo, Yuchen Ma, Zhangchen Xu, Zichen Chen, Nuno Moniz, Zinan Lin, Pin-Yu Chen, Nitesh V. Chawla, Nouha Dziri, Huan Sun, Xiangliang Zhang. Emergent Social Intelligence Risks in Generative Multi-Agent Systems. #2414 Nam Hyeon-Woo, Tae-Hyun Oh, Zeynep Akata, Stephan Alaniz. From Numbers to Narratives: Goal-Oriented Summarization of Machine Learning Model Differences. #2415 Kriti Mahajan. Evolution of Cooperation in LLM Societies : A Multi-Lingual Examination. #2415 Anugyan Das, Omkar Ghugarkar, Vishvesh G Bhat, Asad Aali. Compositional Neuro-Symbolic Reasoning. #2500 Omkar Ghugarkar, Vishvesh G Bhat, Muhammad Ahmed Mohsin, Asad Aali. MAVEN: Improving Generalization in Agentic Tool Calling. #2500 Rajeev Goel, Jason Ding, Phani Harish Wajjala, Pavan K. Turaga, Tejaswi Gowda, Krishna C. Garikipati. CMAG: Concept-Scaffolded Retrieval for Marketplace Avatar Generation. #2501 Ola Aleksandra Pasieka, Dominika Woszczyk, Antoine Cully, Borja G. León. TAME the BALROG: Task-Adaptive Modular Evolution framework for Game Agents. #2501 Shunchang Liu, Xin Chen, Belen Martin Urcelay, Francesco Croce. Preference Instability in Reward Models: Detection and Mitigation via Sparse Autoencoders. #2502 Adarsh Sudheer, David Li, Omar El-Banna, Ishaan Kodarapu, Arjun Bahuguna, Vasu Sharma. Compositional Failure in Audio-Visual LLMs: Late-Layer Prior Dominance Under Cross-modal Conflict. #2502 Abel Jansma. A Compositional Calculus for Semantic Synergy in Language Model Embeddings. #2503 Anand Kamat, Daniel Blake, Brent M. Werness. Grad Detect: Gradient-Based Hallucination Detection in LLMs. #2503 Yuxiao Li, Gjergji Kasneci. Safety Cost of Steering Vectors Is Separable and Reducible. #2504 Carmen Quiles Ramírez, Leticia Lorena Rodriguez, Nicolas Martorell, Natalia Díaz-Rodríguez. Explaining is Harder Than Predicting Alone: Evaluating Concept-based Explanations of MLLMs as ICL Visual Classifiers. #2504 Jiahang Sun, Zhiwei Shang, Zhipiao Liu, Hongwei Yang, XIE GUOQING, Shuang Qiu, Zhongxiang Dai. CB-Orchestrator: Adaptive Workflow Optimization for LLM Agents via Contextual Bandits. #2505 Tianyi Chen, Yinheng Li, Michael Solodko, Sen Wang, Nan Jiang, Junheng Hao, Tingyuan Cui, Jongwoo Ko, Sara Abdali, Suzhen Zheng, Pashmina Cameron, Justin Wagle, Kazuhito Koishida. CUA-Skill: Developing Computer Using Agents with a Skill Framework. #2505 Melissa Wessel. In-Context Learning Amplifies a Latent Compositional Circuit. #2506 Leander Kurscheidt, Gabriele Masina, Roberto Sebastiani, Antonio Vergari. The Theory and Practice of MAP Inference over Non-Convex Constraints. #2506 Prakhar Gupta, Garv Shah, Satyam Goyal, Anirudh Kanchi. Sparse Memory Finetuning as a Low-Forgetting Alternative to LoRA and Full Finetuning. #2507 Sangwoo Cho, Kushal Chawla, Pengshan Cai, Zefang Liu, Chenyang Zhu, Shi-Xiong Zhang, Sambit Sahu. Ask, Don’t Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement. #2507 Lunjun Zhang, Ryan Chen, Bradly C. Stadie. Evolutionary System Prompt Learning for Reinforcement Learning in LLMs. #2508 Ayushi Chadha. When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning. #2508 Hanna Foerster, Tom Blanchard, Kristina Nikolić, Ilia Shumailov, Cheng Zhang, Robert D. Mullins, Nicolas Papernot, Florian Tramér, Yiren Zhao. CaMeLs Can Use Computers Too: System-level Security for Computer Use Agents. #2509 Kairui Zhang, Ziwen Yu, Zahraa S. Abdallah, Martha Lewis. VASAE: Naming SAE Dictionary Directions with Vocabulary-Aligned Anchoring. #2509 Miroslav Lžičář. Successor Re-grounding Audits Compositional Rollout Mismatch in Neuro-Symbolic Search. #2510 Antonin Berthon, Mihaela van der Schaar. Reasoning with Neologisms: Can Soft Tokens Learn Composable Reasoning Skills Without Forgetting?. #2510 Daniil Kirilenko, Dario Fenoglio, Martin Gjoreski, Marc Langheinrich. Visual Counterfactual Explanations with Compositional Generative Models. #2511 Varun Kotte. The Compositional Generalization Gap in Named Entity Recognition: Static Benchmarks Overestimate Transferable Performance. #2511 Mariya I. Vasileva. Large Language Models Can Follow Instructions, But Not Many at Once: Phase Transitions in Compositional Constraint Satisfaction. #2512 Jongseo Lee, Hyuntak Lee, Sunghun Kim, Sooa Kim, Jihoon Chung, Jinwoo Choi. Which Way Did It Move? Diagnosing and Overcoming Directional Motion Blindness in Video LLMs. #2512 Heejeong Nam, Quentin Le Lidec, Lucas Maes, Yann LeCun, Randall Balestriero. Causal-JEPA: Learning World Models through Object-Level Latent Masking. #2312 Hong Kiat Tan, Shariar Kabir, Swastik Agrawal, Sai V R Chereddy, Sriram Balasubramanian. Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries. #2401 Wenhui Huang, Changhe Chen, Han Qi, Chen Lv, Yilun Du, Heng Yang. MoTVLA: A Vision-Language-Action Model with Unified Fast-Slow Reasoning. #2412
12:30 PM	Lunch Break

1:25 PM	Invited Talk: Samy Bengio (Senior Director, Apple and Adjunct Professor, EPFL) Reasoning with LLMs: Challenges and Opportunities.
2:00 PM	Invited Talk: Noémi Éltető (Research Scientist, Google DeepMind) Discovering Interpretable Cognitive Models from Animal and Human Behavior.
2:35 PM	Invited Talk: Francesco Locatello (Assistant Professor, ISTA) Mechanistic interpretability to interpret scientific data.
3:10 PM	Panel Discussion with Noam Brown, Wenya Wang, Noémi Éltető, Aya Abdelsalam Ismail, Parisa Kordjamshidi, and Francesco Locatello
4:00 PM	Poster Session 2 [More] Xuan Gong, Hanbo Huang, Hao Zheng, Yiran Zhang, Wenbin Dai, Weishu Zhao, Shiyu Liang. Reflection Anchors for Interpretable Compositional Visual Reasoning in Multimodal Reinforcement Learning. #2311 Ellen Ma, Nikhil Anand. Irreducible Supervision Enables Compositional Generalization in Post-Training. #2311 Savita Bhat, Vasudeva Varma. What makes the whole? Probing Attribute-Level Compositionality in LLM Judges. #2312 Youssef Zaazou, Mark Thomas. Compositional by Design: Background-Invariant Representations via Linear Additivity in VLMs. #2312 Dhatri C, Tadisetty Sai Yashwanth. Compositional Investigation: Why Reasoning Enables Tool-Using Agents to Fix What They Diagnose. #2313 Apurv Verma, Binh-Nguyen Nguyen, Hai Phan, Lingxiao Wang. Logit Grafting: The Post-Training Delta is Sparse, Portable, and Powerful. #2313 Akshaykumar, Tadisetty Sai Yashwanth. Compositional Skill Acquisition in Agentic Pipelines via Reinforcement Learning and Knowledge Distillation. #2400 Aviral Srivastava, Sourav Panda. Compositional Underdetermination in AI Agents: When Behavioral Success Is Not Compositional Evidence. #2314 Sriram Balasubramanian, Soheil Feizi. Retrieval is Enough: Training-Free Interpretability with a Tool-Using Agent. #2315 Zifan Carl Guo, Laura Ruis, Jacob Andreas, Belinda Z. Li. Introspective Coupling: Self-Explanation Training Tracks Behavioral Change Despite Fixed Supervision. #2315 Vitória Barin-Pacela, Shruti Joshi, Isabela Camacho, Simon Lacoste-Julien, David Klindt. Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisation. #2400 Hakeem Hannoon, Andrew Zhao, Mihir Narayan, Sharvin Goyal, Ivaxi Sheth. Mitigating Over-Personalization in Language Models via Structured Memory. #2400 Anirudh Satheesh, Michael-Andrei Panaitescu-Liess, Andrew Ye Xu, Georgios Milis, Heng Huang, Zikui Cai, Furong Huang. Compositional Adversarial Training for Robust Visual Watermarking. #2401 Piotr Wyrwinski, Kacper Dobek, Krzysztof Krawiec. Structure over Pixels: Learning Variable-Length Visual Programs. #2401 Azwar Abdulsalam, Nishil Patel, Andrew M Saxe. RL Post-Training Builds Compositional Reasoning Strategies. #2402 Nimit Kalra, Narutatsu Ri, Zerzar Bukhari, Ang Li, Sanae Lotfi, Liam H Fowl, Micah Goldblum. Measuring the Limits of Continual Learning for LLMs. #2402 Şuayp Talha Kocabay, Talha Rüzgar Akkuş, Kerem Yalçın. Sample Complexity of Scientific Discovery: PAC Learnability of Compositional Function Trees. #2403 Usman Islam, Zhixun Chen, Stefanos Leonardos; Matteo Leonetti, Yali Du. Policy Transfer for Hierarchical Goal-Conditioned Reinforcement Learning. #2403 Nikhil Garg, Jon Kleinberg, Kenny Peng. How Many Features Can a Language Model Store Under the Linear Representation Hypothesis?. #2404 Kathrin Korte, Christian Medeiros Adriano, Joachim Winther Pedersen, Eleni Nisioti, Sebastian Risi. Dimensionality Controls When Modularity Helps in Continual Learning. #2404 Simon Park, Simran Kaur, Sanjeev Arora. How does RL Post-training Induce Skill Composition? A Case Study on Countdown. #2405 Narutatsu Ri, Abhishek Panigrahi, Sanjeev Arora. Do Thinking Tokens Help with Safety?. #2405 Niclas Dern, Selma Mazioud, Jakob Heiss, Avrajit Ghosh, Curtis James McDonald, Gabriel Clara, Bin Yu. Tracking Training Phases in Compositional Learning with Task-Agnostic Measures. #2406 Luke Cheng, Samuel Lippl. A mathematical theory of balancing relational generalization and memorization. #2406 Wenhao Lu, Spursh Deshpande. What Do Latent Agents Actually Represent? Interpreting Hidden-State Communication in Multi-Agent Systems. #2407 Vikhyath Kothamasu, Virginia Smith, Chhavi Yadav. Hidden in Plain Sight: Benchmarking Agent Safety Against Decomposition Attacks with DeCompBench. #2407 Spursh Deshpande, Wenhao Lu. On the Role of Learned Alignment Matrices in LatentMAS. #2408 Adarsh Kumarappan, Ananya Mujoo. Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy. #2408 Nicole Ching, Ayushi Mehrotra. Sparse Autoencoders Find Causal, Lineage-Specific Context Features in Chromatin Foundation Models. #2409 Pranay Jha. Noise-Tolerant Verification of Compositional Boolean Recovery. #2409 Sebastian Hönig, Su Ji Park, Kushal Jain, Bart Bussmann, Patrick Leask. Additive Relational Bindings in Transformers: What Sparse Autoencoders Miss. #2410 Zijun Wang, Haoqin Tu, Letian Zhang, Hardy Chen, Juncheng Wu, Xiangyan Liu, Zhenlong Yuan, Tianyu Pang, Michael Qizhe Shieh, Fengze Liu, Zeyu Zheng, Huaxiu Yao, Yuyin Zhou, Cihang Xie. Your Agent, Their Asset: A Real-World Safety Analysis of OpenClaw. #2411 Ishaan Kelkar, Nebras Alam, Vikram Kakaria, Madhur Panwar, Vasu Sharma, Maheep Chaudhary. Playing Devil’s Advocate: Off-the-Shelf Persona Vectors Rival Targeted Steering for Sycophancy. #2411 Ryan Co, Karthik Reddy Konuganti. Atomic Chess Reveals Compositional Reasoning Failures in LLMs. #2412 Archer Wang, Emile Timothy Anand, Yilun Du, Marin Soljacic. Unsupervised Decomposition with Recombination-Consistent Diffusion Models. #2412 Cedegao E. Zhang, Cédric Colas, Gabriel Poesia, Joshua B. Tenenbaum, Jacob Andreas. Code-enabled language models can outperform reasoning models on diverse tasks. #2413 Swaminathan S K, Damiya Gondha, Theyanesh Eswaramoorthy Rajahkrishnan, Aritra Hazra. Direction-Conditioned Policies via Compositional Subgoal Scoring for Online Goal-Conditioned Reinforcement Learning. #2414 Jung-Chun Liu, Naihao Deng, Joyce Chai. Language Elicits Compositional Reasoning in Pretrained VFMs. #2414 Jinhan Li, Kexian Tang, Yihan Xu, Zhuorui Ye, Kaifeng Lyu. Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety Reflection. #2415 Juncheng Wu, Letian Zhang, Yuhan Wang, Haoqin Tu, Hardy Chen, Zijun Wang, Cihang Xie, Yuyin Zhou. ClinSeekAgent: Automating Multi-modal Evidence Seeking for Agentic Clinical Reasoning. #2415 Nicholas Bai, Ayushi Mehrotra. Count Me If You Can: Geometric Failure Modes in Language Model Counting. #2500 Zhijin Guo, Richard Freedman, Martha Lewis. From Composition to Compositionality: Discovering Reusable Structure in Polyphonic Music Embeddings. #2500 Clément Castellon, Arindam Biswas. Biregular Sparse Initialization Shifts the Rate and Shape of Compositional Escape in Sequential Arithmetic Curricula. #2501 Shlok Shah, Rhiaan Jhaveri, Tharun Kumar Tiruppali Kalidoss, Chirayu Nimonkar, Ishaan Javali, Dhruv Shah. Emergent Compositional Skills in Mixture-of-Experts VLAs. #2501 Franka Bause, Jonas Niederle, Martin Pawelczyk, Rebekka Burkholz. Multi-Agent Systems are Mixtures of Experts: Who Becomes an Influencer?. #2502 Doojin Baek, Gyubin Lee, Junyeob Baek, Hosung Lee, Sungjin Ahn. Learning to Theorize the World from Observation. #2502 Siyuan Zhang, Jialian Li, Yichi Zhang, Xiao Yang, Yinpeng Dong, Hang Su. Reasoning as State Transition: A Representational Analysis of Reasoning Evolution in Large Language Models. #2503 Heejeong Nam, Chandradithya S Jonnalagadda, Harshit Aggarwal, Eric Xu. Toward Compositional Latent Action Interfaces for Generalizable Agents. #2504 Nicole H. Ma, Nick Rui. Where’s the Plan? Locating Latent Planning in Language Models with Lightweight Mechanistic Interventions. #2504 Sajad Movahedi, Vera Milovanović, Shlomo Libo Feigin, Alexander Theus, Thomas Hofmann, Valentina Boeva, T. Konstantin Rusch, Antonio Orvieto. Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers. #2505 Haozhou Gao, ZENG JIARUI, Wendi Ren, Yanwen Liu, Shuang Li. Dual-Resolution Recursive Energy: Certified Contract–Expand Inference for Sequential Decision Making. #2505 Rishabh Bhattacharya. Unsafe Only in Combination: Interaction-Barrier Shielding for Tool-Using LLM Agents. #2506 Yanwen Liu, Wendi Ren, Haozhou Gao, Shuang Li. Learning What’s Missing: Failure-Driven Skill Discovery via Predicate Bridges. #2506 Kenny Peng, Jon Kleinberg, Nikhil Garg. A Theory of Atomic Features and Four Testable Predictions. #2507 Tilen Gaetano Limbäck-Stokin, Tanishka Birdavade, Kin Ian Lo, Mehrnoosh Sadrzadeh. Meaning Representations as Variational Quantum Circuits. #2507 Quang X. Nguyen, Hieu M. Vu, Dung Viet Nguyen, Hai Tuan Luu, Linh Duy Tran, Tan Minh Nguyen. Separable Representations of Task Complexity and Deliberation in Reasoning Language Models.. #2508 Nirlipta Pande, Georg Niess, Julian Gutheil, Roman Kern, Robert Legenstein, Robert Peharz. HINT: Task Demonstrations for Hierarchical Inference in Abstract Reasoning. #2508 Leo Villani, Sultan Daniels, Ijin Yu, Anant Sahai. Installing and Obstructing Heuristics: Learning Dynamics in Nim. #2509 Heejin Choi. Learning Compositional Tasks via Trigger Compositions: Using Scratchpads as Pre-Answer Workspaces. #2509 Imran Ibrahimli, Stefan Wermter, Jae Hee Lee. Adaptive Recurrence as Algorithmic Time for Length Generalization in Addition. #2510 Samira Abedini, Sina Mavali, Lea Schönherr, Martin Pawelczyk, Rebekka Burkholz. Don't Trust Stubborn Neighbors: A Security Framework for Agentic Networks. #2510 Tianyi Huang, Ming Ren Hou, Jiaheng Su, Yutong Zhang, Ziling Zhang. Compositional Consistency-Guided Decoding for Three-Way Logical Question Answering. #2511 Shuman Peng, Arnas Uselis, Darina Koishigarina, Martin Ester, Seong Joon Oh. CLIP Models Generalize Less Than Compositional Benchmarks Suggest. #2511 Ran Eisenberg, Ofir Lindenbaum. Spatial Compositional Counterfactuals in Concept Bottleneck Models. #2512

Organizers

Sponsors