Mode Collapse

Follow this topic Ask about this topic

Mode collapse occurs when a model concentrates on a few common outputs or solutions while neglecting other valid modes of a distribution, affecting image and video generation, language-model reasoning, robot actions, and Bayesian inference. Approaches include conditioning on modes or intents, distribution matching and replay, regularizing latent-to-output mappings, and changing training noise or sampling to preserve coverage. Researchers assess collapse with measures such as mode coverage, recall, pass@k, and diversity, since single-sample quality or accuracy can hide missing alternatives. The central tension is how to increase diversity without sacrificing quality, and how to tell genuine coverage from merely varied-looking outputs.

Latest papers

34 papers6 papers in the last four weeks, level with the four weeks before. 0.0% of all new papers.

Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data
Apr 20, 2026Zhenwen Liang, Yujun Zhou, Sidi Lu +3Reinforcement Learning RL for Language Model Reasoning
Tencent AI Lab · University of Notre Dame
SubFlow: Sub-mode Conditioned Flow Matching for Diverse One-Step Generation
Apr 14, 2026Yexiong Lin, Jia Shi, Shanshan Ye +3Flow Matching Flow-Based Generative Modeling
Sydney AI Centre, The University of Sydney, Sydney, NSW, Australia · School of Artificial Intelligence, Xidian University, Xi’an, Shaanxi, China · Australian Artificial Intelligence Institute, University of Technology Sydney, Sydney, NSW, Australia +1
Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)
Oct 27, 2025Liwei Jiang, Yuanjun Chai, Margaret Li +7Open-Ended Generation LLM Evaluation
University of Washington · Allen Institute for Artificial Intelligence · Carnegie Mellon University +3
Mode Collapse of Mean-Field Variational Inference
Oct 20, 2025Shunan Sheng, Bohan Wu, Alberto González-SanzVariational Inference Mode Collapse
Columbia University, Department of Statistics · Department of Statistics, Columbia University

Apr 20, 2026J/K move · Enter open · S save

Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data

Zhenwen Liang, Yujun Zhou, Sidi Lu, Xiangliang Zhang+2

Tencent AI Lab · University of Notre Dame

Reinforcement Learning RL for Language Model Reasoning Group Relative Policy Optimization RL Post-Training Mode Collapse

Open paper AskPDF arXiv

Reinforcement Learning (RL) enhances LLM reasoning, yet a paradox emerges as models scale: strong base models saturate standard benchmarks (e.g., MATH), yielding correct but homogeneous solutions. In such environments, the lack of failure cases causes the advantage signal in group-relative algorithms (e.g., GRPO) to vanish, driving policies into mode collapse. To address this, we propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration. Unlike standard sampling that follows model biases, CUTS flattens the local optimization landscape by sampling uniformly from constrained high-confidence candidates. We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance. Experiments on Qwen3 models demonstrate that our approach prevents policy degeneration and significantly boosts out-of-domain generalization. Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO, validating that maintaining diversity within the semantic manifold is critical for rigorous reasoning.

Latest papers

Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data

SubFlow: Sub-mode Conditioned Flow Matching for Diverse One-Step Generation

Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)

Mode Collapse of Mean-Field Variational Inference