Compositional Reasoning

Follow this topic Ask about this topic

Compositional reasoning asks whether AI systems can combine learned pieces into reliable solutions, rather than merely recall facts or match familiar patterns. Approaches include symbolic decomposition and rule chaining, iterative planning and verification, structured neural representations, and training with synthetic examples or reinforcement learning. Researchers test these abilities in settings such as multi-hop question answering, ARC-style visual puzzles, traffic-rule conflicts, and scientific problems requiring combinations of properties. A central tension is that strong performance on individual skills does not guarantee transfer to new compositions, and more inference or training data does not consistently overcome this gap.

Latest papers

39 papers6 papers in the last four weeks, up 50% on the four weeks before. 0.0% of all new papers.

On the Fragility of Tasks under Composition: Implicit Budgeting in Language Models
Oct 19, 2026Chenpan Li, Thies Brockmoeller, Stefanie Jegelka +2Compositional Reasoning Efficient Language Model Reasoning
TU Munich · MIT CSAIL
Recursive Formation and First-Class Reuse of Compositional Capabilities
Oct 14, 2026Hang LinCompositional Reasoning
Reading Right, Acting Wrong: Diagnosing Decision Failures of System One Models in Agent Loops
Oct 13, 2026Yubo Li, Ramayya Krishnan, Rema PadmanLLM Decision-Making LLM Agent Evaluation
Carnegie Mellon University
On-Policy Distillation Teaches New Skills but Not New Knowledge
Oct 7, 2026Yixuan Tang, Yi YangOn-Policy Distillation Knowledge Distillation
The Hong Kong University of Science and Technology
Tropical Reinforcement Learning
Oct 1, 2026Arip Asadulaev, Aladin Djuhera, Karim Salta +3Reinforcement Learning RL for Language Model Reasoning
MBZUAI · TUM
ReHoPER: Receding-Horizon Planning for Enhanced Reasoning
Oct 1, 2026Saeed Ahmadnia, Cornelia CarageaCompositional Reasoning Sequential Decision Making
Department of Computer Science University of Illinois Chicago
A Spectral Theory of Compositional Learning
Sep 27, 2026Hugo RydelComputational Cognitive Modeling Compositional Reasoning
University of Manchester
Compositional Reasoning in Language Models under Reinforcement Learning Post-Training
Sep 16, 2026Yu He, Yingxi Li, Yifei Wang +1Compositional Reasoning RL Post-Training
Stanford University · Amazon AGI Labs
TimeThink: Eliciting Compositional Reasoning in Timeseries Large Language Models
Sep 11, 2026Sudarshan Regmi, Arvind Pillai, Yu Yvonne Wu +7Time Series Reasoning Time Series QA
Dartmouth College
A Multi-Stage Rule-Chaining Framework for Compositional and Interpretable Cognitive Reasoning
Sep 9, 2026Deblina KarCompositional Reasoning Interpretable ML
Indian Institute of Technology Kharagpur
Differentiable Fuzzy Inference Layer: A Monotone, Compositional Ordinal Reasoning Head for Large Language Models
Aug 17, 2026Zhen Zhang, Amr AlanwarCompositional Reasoning Ordinal Regression
School of Computation, Information and Technology Technical University of Munich, Germany
From Atomic Evidence to Logical Composition: Structured Compositional Reasoning over Compound Answer Options
Aug 13, 2026Obed Junias, Maria Leonor PachecoCompositional Reasoning Multiple-Choice Question Answering
University of Colorado Boulder
Geo-Spatial Concept Probing of Large Language Models: Abstraction, Compositionality, and Grounding
Aug 7, 2026Karim Radouane, Jose G Moreno, Lynda TamineSpatial Reasoning Benchmarks LLM Grounding
University of Toulouse, IRIT Toulouse, France
The Ignition Is Real, and It Lives at the Readout: Latent composition, difficulty-clocked ignition, and the interface-constituted commit in a recurrent-depth reasoner
Aug 4, 2026Simon Lam-MuirCompositional Reasoning Representation Geometry in Language Models
Prime Calibre, Australia
RECON: Benchmarking Agent Memory for Compositional Reasoning over Long Contexts
Jul 18, 2026Mihir Shriniwas AryaCompositional Reasoning LLM Agent Memory
Department of Computer Science and Engineering RV College of Engineering
Concretized Proposition Prompting Resolves Composition-Knowledge Dichotomy in Large Language Models
Jul 9, 2026Changhun Lee, Minguk Jeon, Jongkyung Shin +1Compositional Reasoning LLM Prompting
Columbia University · UNIST · POSCO Holdings
RL Post-Training Builds Compositional Reasoning Strategies
Jul 8, 2026Azwar Abdulsalam, Nishil Patel, Andrew SaxeCompositional Reasoning RL for Language Model Reasoning
Gatsby Computational Neuroscience Unit, UCL, London, United Kingdom.
ReasonCLIP-58M: Visually Grounded Commonsense Reasoning Supervision for CLIP
Jun 25, 2026Sicheng Zhang, Muzammal Naseer, Binzhu Xie +5Compositional Reasoning Multimodal Reasoning
Khalifa University · University of Western Australia · The Chinese University of Hong Kong +2
Holographic Memory for Zero-Shot Compositional Reasoning in Knowledge Graphs: A Mechanistic Study of Where and Why It Fails
Jun 23, 2026Randhir KumarCompositional Reasoning KG Embedding
Independent Researcher
Mat-Pref: Verifiable-Reward Training Improves Compositional Reasoning in Inorganic Materials
Jun 20, 2026Sarrah R. Mikhail Leung, Taehan Kim, Jeongbin ParkCompositional Reasoning RL for Language Model Reasoning
University of California, Berkeley · 2Machine Learning at Berkeley · University of Michigan
Compositional Reasoning Depth Predicts Clinical AI Failure: Empirical Evidence Consistent with Transformer Compositionality Limits in Electronic Health Record Question Answering
Jun 15, 2026Sanjay BasuMulti-Hop QA Compositional Reasoning
Department of Medicine, University of California San Francisco, San Francisco, CA · Waymark, San Francisco, CA
Operadic consistency: a label-free signal for compositional reasoning failures in LLMs
Jun 11, 2026Nathaniel Bottman, Yinhong Liu, Kyle RichardsonCompositional Reasoning Reasoning Consistency in Language Models
Incubilate · University of Cambridge · Allen Institute for Artificial Intelligence
Operads for compositional reasoning in LLMs
Jun 11, 2026Nathaniel Bottman, Kyle RichardsonMulti-Hop QA Compositional Reasoning
Incubilate, Seattle, USA · Allen Institute for Artificial Intelligence, Seattle, USA
Principles of Concept Representation in Sentence Encoders
Jun 5, 2026Isabelle Mohr, John Dujany, Jonathan Souquet +1Sentence Embeddings Compositional Reasoning
Idiap Research Institute · Merck KGaA
R-APS: Compositional Reasoning and In-Context Meta-Learning for Constrained Design via Reflective Adversarial Pareto Search
Jun 3, 2026João Pedro Gandarela, Thiago Rios, Stefan Menzel +1Mechanism Design Meta-Learning
Idiap Research Institute, Switzerland · École Polytechnique Fédérale de Lausanne (EPFL), Switzerland · Honda Research Institute Europe, Germany +2
Composition Collapse: Stable Factual Knowledge Does Not Imply Compositional Reasoning
May 26, 2026Zhe Yu, Wenpeng Xing, Yunzhao Wei +4Compositional Reasoning Reasoning Evaluation
Binjiang Institute of Zhejiang University · Zhejiang University · Hong Kong Baptist University +2
Convex Compositional Reasoning Models
May 22, 2026Meir Roketlishvili, Semyon Semenov, Maksim Bobrin +7Compositional Reasoning Energy-Based Models
Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE · Applied AI Institute, Computational Imaging Lab · AXXX
Shortcut Solutions Learned by Transformers Impair Continual Compositional Reasoning
May 6, 2026William T. Redman, Erik C. Johnson, Brian RobinsonCompositional Reasoning Continual Learning
Johns Hopkins Applied Physics Lab
The Power of Power Law: Asymmetry Enables Compositional Reasoning
Apr 24, 2026Zixuan Wang, Xingyu Dang, Jason D. Lee +1Long-Tail Learning Compositional Reasoning
1Princeton University · University of California, Berke · 3Tsinghua University
KnowledgeBerg: Evaluating Systematic Knowledge Coverage and Compositional Reasoning in Large Language Models
Apr 19, 2026Xiao Zhang, Qianru Meng, Yongjian Chen +2LLM Evaluation Compositional Reasoning
University of Groningen · LIACS, Leiden University

Oct 19, 2026J/K move · Enter open · S save

On the Fragility of Tasks under Composition: Implicit Budgeting in Language Models

Chenpan Li, Thies Brockmoeller, Stefanie Jegelka, Sharut Gupta+1

TU Munich · MIT CSAIL

Compositional Reasoning Efficient Language Model Reasoning Adaptive Reasoning in Language Models

Open paper AskPDF arXiv

Modern language models are increasingly asked to handle several tasks within a single prompt. Doing so requires solving each task correctly while integrating the resulting answers into a coherent response. In this paper, we identify a fundamental limitation in this process, where models make substantially more errors when tasks are composed, even when they can solve every constituent task reliably in isolation. This degradation cannot be fully explained by the accumulation of independent errors. Across both open-weight models and the frontier model Claude Opus 5, our findings instead point to an implicit thinking budget underlying this effect. In particular, under composition, models use substantially fewer tokens per task than in isolation, even when the full sequence remains well within the context window. Tasks that require longer solution traces in isolation are consequently more

\textit{fragile}

under composition, i.e., models make more errors on prompts containing these tasks. In isolation, non-fragile tasks receive more reasoning than they need and waste compute. However, by grouping tasks according to their fragility, the implicit budgeting can be manipulated to reduce token usage by

43

% across a range of math and linguistic tasks while also preserving accuracy.