Reinforcement Learning from Denoising Feedback
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910254719041536 |
|---|---|
| author | He, Qi Chen, Huan Guo, Ya Zhu, Huijia Fung, Yi R. Zhou, Baojian |
| author_facet | He, Qi Chen, Huan Guo, Ya Zhu, Huijia Fung, Yi R. Zhou, Baojian |
| contents | Policy loss estimation remains a fundamental and long-standing challenge in reinforcement learning (RL) for diffusion language models (dLLMs). We introduce Reinforcement Learning from Denoising Feedback (RLDF), a novel training paradigm that leverages feedback obtained from rollout and training processes to facilitate accurate and efficient policy loss estimation. To balance the trade-off between computational efficiency and estimation effectiveness, RLDF optimizes the model toward the clipped clean state $\hat{x}_0$ from intermediate noisy states $x_t$, combined with weighted timestep sampling over $t$. Extensive experiments demonstrate that RLDF achieves consistent and substantial improvements in both performance and generalizability across two representative dLLM architectures, LLaDA and Dream, on multiple reasoning benchmarks. Our work lays a principled foundation for scalable reinforcement learning in diffusion language models. We build Drift, a training framework for dLLMs, available at https://github.com/ant-research/Drift. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_25638 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Reinforcement Learning from Denoising Feedback He, Qi Chen, Huan Guo, Ya Zhu, Huijia Fung, Yi R. Zhou, Baojian Computation and Language Machine Learning Policy loss estimation remains a fundamental and long-standing challenge in reinforcement learning (RL) for diffusion language models (dLLMs). We introduce Reinforcement Learning from Denoising Feedback (RLDF), a novel training paradigm that leverages feedback obtained from rollout and training processes to facilitate accurate and efficient policy loss estimation. To balance the trade-off between computational efficiency and estimation effectiveness, RLDF optimizes the model toward the clipped clean state $\hat{x}_0$ from intermediate noisy states $x_t$, combined with weighted timestep sampling over $t$. Extensive experiments demonstrate that RLDF achieves consistent and substantial improvements in both performance and generalizability across two representative dLLM architectures, LLaDA and Dream, on multiple reasoning benchmarks. Our work lays a principled foundation for scalable reinforcement learning in diffusion language models. We build Drift, a training framework for dLLMs, available at https://github.com/ant-research/Drift. |
| title | Reinforcement Learning from Denoising Feedback |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2605.25638 |