AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Gradient Regularization Mitigates Reward Hacking in Reinforcement Learning from Human Feedback and Verifiable Rewards

Johannes Ackermann, Michael Noukhovitch, Takashi Ishida, Masashi Sugiyama

ICML 2026 · Reinforcement learning › RL from feedback · approach: Theory of RLHF · 0 citations (OpenAlex)

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
add Reference Resets of the KL reference to GRPO (ablation)GRPO with KL penalty · TL;DR win rate, Pythia 1Bwin rate: 62.2% -> 78.1%Improved
add Reference Resets to GRPO (ablation)GRPO with KL penalty · TL;DR win rate, Qwen2.5-1.5Bwin rate: 82.5% -> 82.0%No clear effect
add Reference Resets to DPO (TR-DPO)DPO · TL;DR win ratewin rate: 45.9% -> 45.5% (Pythia 1B)No clear effect
lower KL penalty strength beta instead of using Reference Resets (ablation)GRPO with KL penalty · TL;DR, Pythia 1Bgold model scoreHurt
replace KL penalty with explicit gradient regularization (ablation)GRPO · AlpacaFarm, Qwen2.5-3Bwin rate: 52.8% -> 59.2%Improved
add KL penalty to unregularized RLHF (ablation)GRPO · AlpacaFarm, Qwen2.5-3Bwin rate: 44.2% -> 52.8%Improved
use GR instead of Reference Resets with the strongest RMGRPO · AlpacaFarm, Qwen2.5-1.5B, strongest RMwin rateHurt
add GR to GRPO with rule-based reward (ablation)GRPO · GSM8K, Qwen2.5-0.5Baccuracy: 52.1 ± 0.3% -> 56.2 ± 1.0%Improved
add KL penalty to GRPO with rule-based reward (ablation)GRPO · GSM8K, Qwen2.5-0.5Baccuracy: 52.1 ± 0.3% -> 44.2 ± 1.2%Hurt
add GR to GRPO with LLM-judge reward (ablation)GRPO · GSM8K, Qwen2.5-1.5Baccuracy: 1.9% -> 67.8%Improved
add GR to GRPO with LLM-judge reward (ablation)GRPO · GSM8K, Qwen2.5-0.5Baccuracy: 20.7 ± 3.8% -> 41.0 ± 2.8%Improved
add GR to GRPO with rule-based reward (ablation)GRPO · MATH, Qwen2.5-1.5B-Instructpass@1: 54.8% -> 57.6%Improved

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