RL Post-Training Updates Can Be Sparsely Decomposed
menhguin · x · 2026-07-15
This repost covers new research on RL post-training, addressing the core issue: while RL updates are effective, the parameter changes themselves act as a "black box."
The paper proposes treating the genuinely effective "reasoning component" of RL as a compact reconnection matrix within the base model's spectral space. Based on this, they introduce SAR: a retraining-free post-processing method that projects raw RL updates onto this "reasoning core" to better understand, purify, and merge RL-trained models.
Key conclusions from the text include:
- The reasoning core is highly sparse: Less than 1% of the spectral parameters are needed to recover or even improve upon the full RL gains.
- Capable of denoising: Removing irrelevant/noisy directions maintains or enhances model performance.
- Useful for model merging: Beyond analyzing RL updates, it allows post-training modifications to be integrated back into models more cleanly.
More from Research
- LoMa Paper Ships REALLY HardPairs Dataset, Accepted at ECCV 2026 — ducha_aiki · 2026-09-11
- Johns Hopkins Launches Full-Stack Hands-on Robot Learning Class with SO-101 Arm Kits — _krishna_murthy · 2026-09-11
- SyncWorld: In-Context Robot World Model Simulates Unseen Views and Embodiments Zero-Shot — ChongZzZhang · 2026-09-11
- A 3D Pose Dataset for Dogs Released — ducha_aiki · 2026-09-11
- Five tells that still make AI video read as AI, from physics glitches to missing operators — NewPhoneWhotiz · 2026-09-11
- AnyMatch accepted to ECCV 2026 with a NoPresenter design — ducha_aiki · 2026-09-11