Papers

Filtered to verifiable rewards · clear filter

Browse by term

continual learning 33reinforcement learning 21large language models 9vision-language models 8language models 6benchmarking 5autoregressive models 4diffusion transformers 4generative models 4multimodal learning 4robotics 4video generation 4benchmarks 3computer vision 3policy optimization 3self-distillation 3vision-language-action models 3world modeling 3agent-based systems 2autonomous agents 2calibration 2coding agents 2diffusion models 2foundation models 2image synthesis 2in-context learning 2knowledge graphs 2LLMs 2multimodal evaluation 2multimodal large language models 2

Matching papers

ISO: An RLVR-Native Optimization Stack

5 upvotes · 21 JUL 2026 · Hanqing Zhu, Wenyan Cong, Zhizhou Sha et al.

This paper develops a new optimization framework called Isospectral Optimization (ISO) that improves the performance of language models trained using reinforcement learning with verifiable rewards (RLVR) by reusing the base model's weight spectra while adapting the input and output singular frames. Practitioners might care about this because it could lead to faster and more accurate training of large language models.

H^2SD: Hybrid Hindsight Self-Distillation

5 upvotes · 21 JUL 2026 · Qiye Cai, Yichuan Ma, Linyang Li et al.

This paper introduces H^2SD, a hybrid hindsight self-distillation framework for reinforcement learning with verifiable rewards, which combines the strengths of different methods to improve large language models' reasoning capabilities. Practitioners may care about this work because it addresses limitations of existing methods and shows promising results on challenging reasoning benchmarks.