The Path Not Taken: RLVR Provably Learns Off the Principals
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914151485407232 |
|---|---|
| author | Zhu, Hanqing Zhang, Zhenyu Huang, Hanxian Su, DiJia Liu, Zechun Zhao, Jiawei Fedorov, Igor Pirsiavash, Hamed Sha, Zhizhou Lee, Jinwon Pan, David Z. Wang, Zhangyang Tian, Yuandong Tai, Kai Sheng |
| author_facet | Zhu, Hanqing Zhang, Zhenyu Huang, Hanxian Su, DiJia Liu, Zechun Zhao, Jiawei Fedorov, Igor Pirsiavash, Hamed Sha, Zhizhou Lee, Jinwon Pan, David Z. Wang, Zhangyang Tian, Yuandong Tai, Kai Sheng |
| contents | Reinforcement Learning with Verifiable Rewards (RLVR) reliably improves the reasoning performance of large language models, yet it appears to modify only a small fraction of parameters. We revisit this paradox and show that sparsity is a surface artifact of a model-conditioned optimization bias: for a fixed pretrained model, updates consistently localize to preferred parameter regions, highly consistent across runs and largely invariant to datasets and RL recipes. We mechanistically explain these dynamics with a Three-Gate Theory: Gate I (KL Anchor) imposes a KL-constrained update; Gate II (Model Geometry) steers the step off principal directions into low-curvature, spectrum-preserving subspaces; and Gate III (Precision) hides micro-updates in non-preferred regions, making the off-principal bias appear as sparsity. We then validate this theory and, for the first time, provide a parameter-level characterization of RLVR's learning dynamics: RLVR learns off principal directions in weight space, achieving gains via minimal spectral drift, reduced principal-subspace rotation, and off-principal update alignment. In contrast, SFT targets principal weights, distorts the spectrum, and even lags RLVR.
Together, these results provide the first parameter-space account of RLVR's training dynamics, revealing clear regularities in how parameters evolve. Crucially, we show that RL operates in a distinct optimization regime from SFT, so directly adapting SFT-era parameter-efficient fine-tuning (PEFT) methods can be flawed, as evidenced by our case studies on advanced sparse fine-tuning and LoRA variants. We hope this work charts a path toward a white-box understanding of RLVR and the design of geometry-aware, RLVR-native learning algorithms, rather than repurposed SFT-era heuristics. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_08567 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | The Path Not Taken: RLVR Provably Learns Off the Principals Zhu, Hanqing Zhang, Zhenyu Huang, Hanxian Su, DiJia Liu, Zechun Zhao, Jiawei Fedorov, Igor Pirsiavash, Hamed Sha, Zhizhou Lee, Jinwon Pan, David Z. Wang, Zhangyang Tian, Yuandong Tai, Kai Sheng Machine Learning Artificial Intelligence Reinforcement Learning with Verifiable Rewards (RLVR) reliably improves the reasoning performance of large language models, yet it appears to modify only a small fraction of parameters. We revisit this paradox and show that sparsity is a surface artifact of a model-conditioned optimization bias: for a fixed pretrained model, updates consistently localize to preferred parameter regions, highly consistent across runs and largely invariant to datasets and RL recipes. We mechanistically explain these dynamics with a Three-Gate Theory: Gate I (KL Anchor) imposes a KL-constrained update; Gate II (Model Geometry) steers the step off principal directions into low-curvature, spectrum-preserving subspaces; and Gate III (Precision) hides micro-updates in non-preferred regions, making the off-principal bias appear as sparsity. We then validate this theory and, for the first time, provide a parameter-level characterization of RLVR's learning dynamics: RLVR learns off principal directions in weight space, achieving gains via minimal spectral drift, reduced principal-subspace rotation, and off-principal update alignment. In contrast, SFT targets principal weights, distorts the spectrum, and even lags RLVR. Together, these results provide the first parameter-space account of RLVR's training dynamics, revealing clear regularities in how parameters evolve. Crucially, we show that RL operates in a distinct optimization regime from SFT, so directly adapting SFT-era parameter-efficient fine-tuning (PEFT) methods can be flawed, as evidenced by our case studies on advanced sparse fine-tuning and LoRA variants. We hope this work charts a path toward a white-box understanding of RLVR and the design of geometry-aware, RLVR-native learning algorithms, rather than repurposed SFT-era heuristics. |
| title | The Path Not Taken: RLVR Provably Learns Off the Principals |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2511.08567 |