Compositional Generalization

Follow this topic Ask about this topic

Compositional generalization studies whether models can reuse learned concepts, skills, and rules in combinations they did not encounter during training. Approaches include modular and object-centric architectures, factorized or symbolic representations, and training methods such as curricula, structured reasoning traces, and reinforcement learning. Researchers test these abilities on controlled language, vision, arithmetic, planning, and robotics tasks, often holding out combinations, deeper structures, or longer sequences. Results remain uneven: models often recombine familiar parts more reliably than they discover new ones, and strong test accuracy can reflect shortcuts rather than reusable structure.

Latest papers

113 papers12 papers in the last four weeks, up 9% on the four weeks before. 0.1% of all new papers.

Known Intents, New Combinations: Clause-Factorized Decoding for Compositional Multi-Intent Detection
Mar 30, 2026Abhilash NandyIntent Classification Compositional Generalization
Microsoft Research India
Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisation
Mar 30, 2026Vitória Barin-Pacela, Shruti Joshi, Isabela Camacho +2Sparse Autoencoders Sparse Coding
Mila - Québec AI Institute & Université de Montréal · Santa Clara University · Cold Spring Harbor Laboratory
Arithmetic OOD Failure Unfolds in Stages in Minimal GPTs
Mar 27, 2026Seine A. ShintaniNumerical Reasoning in Language Models OOD Generalization
Department of Biomedical Sciences, College of Life and Health Sciences, Chubu University, Kasugai, Aichi 487-8501, Japan · Center for Mathematical Science and Artificial Intelligence, Chubu University, Kasugai, Aichi 487-8501, Japan · Institute for Advanced Research, Nagoya University, Nagoya, Aichi 464-8601, Japan
Towards Generalizable Robotic Data Flywheel: High-Dimensional Factorization and Composition
Mar 26, 2026Yuyang Xiao, Yifei Zhou, Haoran Wang +2Data-Efficient Robot Learning Robotic Data Collection
ByteDance Seed
Thinking Deeper, Not Longer: Memory-Efficient Test-Time Reasoning with Depth-Recurrent Transformers for Compositional Generalization
Mar 23, 2026Hung-Hsuan ChenEfficient Transformer Inference Recurrent Transformers
Computer Science and Information Engineering National Central University Taoyuan, Taiwan
How Psychological Learning Paradigms Shaped and Constrained Artificial Intelligence
Mar 18, 2026Alex Anvi Eponon, Ildar Batyrshin, Christian E. Maldonado-Sifuentes +1Cognitive Architectures for AI Agents Compositional Generalization
Centro de Investigación en Computación (CIC), Instituto Politécnico Nacional (IPN), Mexico City, Mexico.
Functorial Neural Architectures from Higher Inductive Types
Mar 17, 2026Karen SargsyanCompositional Generalization Compositional Representation Learning
Institute of Chemistry, Academia Sinica, Taipei, Taiwan
Ternary Gamma Semirings: From Neural Implementation to Categorical Foundations
Mar 15, 2026Ruoqi SunNeural Network Generalization Compositional Representation Learning
Intelligent Game and Decision Lab(IGDL), No.1 Xianghongqi Road, Haidian District, Beijing, 100091,China.
The ARC of Progress towards AGI: A Living Survey of Abstraction and Reasoning
Mar 9, 2026Sahar Vahdati, Andrei Aioanei, Haridhra Suresh +1Abstract Reasoning ARC-AGI
TIB - Leibniz Information Centre for Science and Technology · L3S Research Center, Leibniz University of Hannover · Dresden University of Technology +1
Compositional Generalization Requires Linear, Orthogonal Representations in Vision Embedding Models
Feb 27, 2026Arnas Uselis, Andrea Dittadi, Seong Joon OhNeural Representation Geometry Visual Representation Learning
Tübingen AI Center, University of Tübingen · Helmholtz · Technical University of Munich
COMiT: Learning Structured Visual Tokens through Sequential Communication
Feb 24, 2026Aram Davtyan, Yusuf Sahin, Yasaman Haghighi +4Image Tokenization Visual Tokenization
Computer Vision Group, University of Bern, Switzerland · VITA Lab, EPFL, Switzerland
Neuro-Symbolic Decoding of Neural Activity
Feb 22, 2026Yanchen Wang, Joy Hsu, Ehsan Adeli +1Neural Decoding fMRI-Based Visual Decoding
Columbia University · Stanford University
Are Object-Centric Representations Better At Compositional Generalization?
Feb 18, 2026Ferdinand Kapl, Amir Mohammad Karimi Mamaghan, Maximilian Seitzer +4Compositional Generalization Object-Centric Representation Learning
Technical University of Munich · Helmholtz AI, Munich · KTH Royal Institute of Technology +2
Why Can't I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action Recognition
Jan 22, 2026Geo Ahn, Inwoong Lee, Taeoh Kim +3Zero-Shot Learning Video Action Recognition
1Kyung Hee University · ∗This work was done during an internship at NAVER Cloud. · 2NAVER Cloud
Is Generation Required for Data-Efficient Perception?
Dec 9, 2025Jack Brady, Bernhard Schölkopf, Thomas Kipf +2Visual Representation Learning Generative Modeling
Max Planck Institute for Intelligent Systems, Tübingen, Germany · Tübingen AI Center · ELLIS Institute, Tübingen +1
How Does RL Post-training Induce Skill Composition? A Case Study on Countdown
Dec 1, 2025Simon Park, Simran Kaur, Sanjeev AroraCompositional Generalization RL Post-Training
Princeton Language and Intelligence (PLI), Princeton University
Understanding the Staged Dynamics of Transformers in Learning Latent Structure
Nov 24, 2025Rohan Saha, Farzane Aminmansour, Alona FysheTransformer Training Dynamics Decoder-Only Language Models
Department of Computing Science University of Alberta Edmonton, Canada · Department of Computing Science and Psychology University of Alberta Edmonton, Canada
Memorizing Long-tail Data Can Help Generalization Through Composition
Oct 18, 2025Mo Zhou, Haoyang Ma, Rong GeNeural Network Generalization Neural Network Memorization
University of Washington · Duke University
What Drives Compositional Generalization in Visual Generative Models? The Importance of Continuous Training Objectives
Oct 3, 2025Karim Farid, Rajat Sahay, Yumna Ali Alnaggar +4Compositional Image Generation Generative Modeling
University of Freiburg · Bosch Center for Artificial Intelligence · Inria, École Normale Supérieure, CNRS, PSL Research University
Unveiling the Mechanisms of Multi-Hop Reasoning in Transformers via Identity Bridge
Sep 29, 2025Pengxiao Lin, Zheng-An Chen, Zhi-Qin John XuTransformer Interpretability OOD Generalization
School of Mathematical Sciences, Shanghai Jiao Tong University · Institute of Natural Sciences, MOE-LSC, Shanghai Jiao Tong University · Shanghai Seres Information Technology Co., Ltd, Shanghai 200040, China.
Learning Composable Chains-of-Thought
May 28, 2025Fangcong Yin, Zeyu Leo Liu, Liu Leqi +2LLM Fine-Tuning CoT Reasoning
The University of Texas at Austin · Princeton University
Disassociating performance from compositional feature learning
May 14, 2025George Dimitriadis, Spyridon SamothrakisRepresentation Learning OOD Generalization
Computer Science and Electronic Engineering Department of the University of Essex
A Theoretical Analysis of Provable Compositional Generalization in Neural Networks: A Necessary and Sufficient Condition
May 5, 2025Yuanpeng LiCompositional Generalization

Mar 30, 2026J/K move · Enter open · S save

Known Intents, New Combinations: Clause-Factorized Decoding for Compositional Multi-Intent Detection

Abhilash Nandy

Microsoft Research India

Intent Classification Compositional Generalization Multi-Label Classification

Open paper AskPDF arXiv

Multi-intent detection papers usually ask whether a model can recover multiple intents from one utterance. We ask a harder and, for deployment, more useful question: can it recover new combinations of familiar intents? Existing benchmarks only weakly test this, because train and test often share the same broad co-occurrence patterns. We introduce CoMIX-Shift, a controlled benchmark built to stress compositional generalization in multi-intent detection through held-out intent pairs, discourse-pattern shift, longer and noisier wrappers, held-out clause templates, and zero-shot triples. We also present ClauseCompose, a lightweight decoder trained only on singleton intents, and compare it to whole-utterance baselines including a fine-tuned tiny BERT model. Across three random seeds, ClauseCompose reaches 95.7 exact match on unseen intent pairs, 93.9 on discourse-shifted pairs, 62.5 on longer/noisier pairs, 49.8 on held-out templates, and 91.1 on unseen triples. WholeMultiLabel reaches 81.4, 55.7, 18.8, 15.5, and 0.0; the BERT baseline reaches 91.5, 77.6, 48.9, 11.0, and 0.0. We also add a 240-example manually authored SNIPS-style compositional set with five held-out pairs; there, ClauseCompose reaches 97.5 exact match on unseen pairs and 86.7 under connector shift, compared with 41.3 and 10.4 for WholeMultiLabel. The results suggest that multi-intent detection needs more compositional evaluation, and that simple factorization goes surprisingly far once evaluation asks for it.