AI Paper Atlas

Research map / Reinforcement learning / Offline RL

RL's Razor: Why Online Reinforcement Learning Forgets Less

Idan Shenfeld, Jyothish Pari, Pulkit Agrawal

ICLR 2026 · Reinforcement learning › Offline RL · approach: Offline-to-online fine-tuning

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune with RL (GRPO) instead of SFT at matched new-task performanceQwen2.5-3B-Instruct, OpenVLA-7B · Math, Science Q&A, Tool use, robot pick-and-placeprior-task retention vs new-task accuracyImproved
use forward KL on new task as predictor of forgetting (ablation)ParityMNIST MLP · ParityMNIST/FashionMNISTR^2 quadratic fit: 0.96 (forward KL) vs 0.34-0.93 for alternativesImproved
forward KL as predictor in LLM experimentsQwen2.5-3B-Instruct · LLM tasksR^2: 0.96 (MNIST) -> 0.71 (LLM)Mixed
train SFT on oracle minimum-KL distribution instead of arbitrary labels (ablation)SFT on ParityMNIST · ParityMNIST/FashionMNISTprior-task retention at matched accuracyImproved
distill SFT from data generated by an RL-trained model (ablation)SFT on ParityMNIST · ParityMNIST/FashionMNISTaccuracy-forgetting trade-offImproved
use on-policy data without negative examples (1-0 Reinforce) instead of SFT (ablation)Qwen2.5-3B-Instruct · Science Q&Aforgetting and KL shiftImproved
add negative examples to offline training (SimPO vs SFT) (ablation)offline objective · Science Q&AforgettingNo clear effect
use float32 instead of bfloat16 for RL training (ablation)RL · weight update sparsityupdate sparsityNo clear effect
increase model size from 3B to 14B under SFT (ablation)Qwen2.5 SFT · Science Q&Alearning-forgetting trade-offNo clear effect
fine-tune with RL instead of SFTQwen2.5-3B-Instruct · Wikipedia representation similarityCKNNA to base: 0.56 -> 0.94Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy