Regularized Score Centering fixes advantage estimation in agentic RL, 9.9× on BabyAI
aviral_kumar2 · x · 2026-11-20
jackbotcs introduces Regularized Score Centering (RSC): classic Score Centering subtracts the sampler's mean score per token but fails when the learner is far from the sampler; RSC adds a reverse KL to the sampler as a regularizer. RSC achieves the best results on all six agentic tasks, including a 9.9× improvement on BabyAI. Details in a 13-post thread.
Related event: Regularized Score Centering Boosts Agentic RL Performance(2 posts)→
More from Research
- Testing browser agents on the live web is unsafe, so this team built a hermetic synthetic "Mirror Dimension" — laparisa · 2026-11-20
- Road MOPD: resolving multi-teacher distillation conflicts, +5.6–8.8 pts on medical tasks — jiqizhixin · 2026-11-20
- Roboticist argues full joint space is the wrong abstraction for high-level agents — KyleMorgenstein · 2026-11-20
- Do language models have explicit goal encodings? New interpretability paper takes first step — benno_krojer · 2026-11-20
- DeepMind Adds Memory to Recursive Self-Improvement, Boosting Scores 68→85 — Dr_Singularity · 2026-11-20
- LLM Malaria Drug Discovery Paper Accepted at NeurIPS Africa in AI Workshop — madeofAjala · 2026-11-20