arXiv AI

Spurious Advantage Hidden in GRPO

Group Relative Policy Optimization (GRPO) assigns a magnitude to each rollout based on within‑group reward statistics, rewarding rollouts that reach the correct answer through reasoning. However, the same magnitude can be high for rollouts that reach the answer by guessing, creating a spurious advantage that misleads the policy toward guess‑like behaviors. The paper identifies three scenarios where this occurs—bounded‑answer tasks, open‑answer sets with bounded sub‑cases, and search agents with many paths to the same answer—and proposes SIGNBALANCE, a composition‑free magnitude that preserves the verifier sign, uses a global scale, and restores zero‑mean balance via stop‑gradient per‑class rescaling, matching GRPO on open‑answer math and improving on bounded‑answer math and search agents.

Hugging Face Trending Papers
Sep 3

Spurious Advantage Hidden in GRPO

The paper identifies a flaw in Group Relative Policy Optimization (GRPO) called the spurious advantage, where the advantage estimator can incorrectly reward rollouts that reach correct answers by guessing rather than reasoning. This issue appears in bounded-answer tasks, open-answer sets with bounded sub-cases, and search agents with many paths to the same answer. The authors propose SIGNBALANCE, a new magnitude estimator that removes composition, maintains verifier sign, applies a global scale, and uses stop‑gradient per‑class rescaling to restore zero‑mean balance, achieving comparable or better performance on math and search benchmarks.

arXiv AI
Sep 4

F-GRPO: Don't Let Your Policy Learn the Obvious and Forget the Rare

The paper introduces F-GRPO, a method that addresses the issue of reinforcement learning policies overfitting to common trajectories while neglecting rare correct ones. By deriving the probability of prompt‑local tail‑miss events and proposing a difficulty‑aware scaling coefficient inspired by Focal loss, the authors show that down‑weighting high‑success sampled groups can improve performance. Experiments on categorical simulations, Maze, and large language models (Qwen2.5‑7B) demonstrate that F‑GRPO raises average math pass rates and out‑of‑distribution performance without increasing group size or computational cost.

By Daniil Plyusov, Alexey Gorbatovski, Boris Shaposhnikov, Viacheslav Sinii, Alexey Malakhov, Daria Korotyshova, Daniil Gavrilov
arXiv Machine Learning
Aug 24

Reinforcing Multi-Turn Reasoning in LLM Agents via Fine-Grained Reward Structure and Credit Assignment

The paper explores how dense, turn-level reward structures can improve reinforcement learning for large language model agents in multi-turn tasks. It introduces three reward granularity types—terminal, delayed, and per-turn—and adapts Group Relative Policy Optimization and Proximal Policy Optimization to each. Experiments on search and game agents show that per-turn rewards consistently yield better training dynamics, faster convergence, and higher answer correctness compared to sparse terminal or delayed rewards.

By Quan Wei, Siliang Zeng, Chenliang Li, Zhongruo Wang, William Brown, Oana Frunza, Wei Deng, Anderson Schneider, Yuriy Nevmyvaka, Yang Katie Zhao, Alfredo Garcia, Mingyi Hong
arXiv Machine Learning
Jun 25

ExTra: Exploratory Trajectory Optimization for Language Model Reinforcement Learning

arXiv:2606. 24994v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) for language-model reasoning can fail at both extremes of task difficulty: easy prompts often produce all-correct, low-diversity rollout groups with little gradient signal, while hard prompts can produce all-incorrect groups with no positive reward.

By Wenyang Hu, Junxiang Jia, Zhen Shu, Daniel Dahlmeier, See-Kiong Ng, Bryan Kian Hsiang Low
arXiv Machine Learning
Jun 16

DRA-GRPO: Your GRPO Needs to Know Diverse Reasoning Paths for Mathematical Reasoning

arXiv:2505. 09655v5 Announce Type: replace-cross Abstract: Post-training LLMs with Reinforcement Learning, specifically Group Relative Policy Optimization (GRPO), has emerged as a paradigm for enhancing mathematical reasoning.

By Xiwen Chen, Wenhui Zhu, Peijie Qiu, Xuanzhao Dong, Hao Wang, Haiyu Wu, Huayu Li, Aristeidis Sotiras, Yalin Wang, Abolfazl Razi