Shrinking the Variance: Shrinkage Baselines for Reinforcement Learning with Verifiable Rewards

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zeng, Guanning, Zhou, Zhaoyi, Arora, Daman, Zanette, Andrea
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914337237499904
author Zeng, Guanning
Zhou, Zhaoyi
Arora, Daman
Zanette, Andrea
author_facet Zeng, Guanning
Zhou, Zhaoyi
Arora, Daman
Zanette, Andrea
contents Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a powerful paradigm for post-training large reasoning models (LRMs) using policy-gradient methods such as GRPO. To stabilize training, these methods typically center trajectory rewards by subtracting the empirical mean reward for each prompt. Statistically, this centering acts as a control variate (baseline), reducing the variance of the policy-gradient estimator. In practice, the mean reward is estimated using per-prompt empirical averages computed from the generations for each prompt in a batch. Motivated by Stein's paradox, we propose shrinkage estimators that combine per-prompt and across-prompt means to improve per-prompt mean estimation accuracy, especially in the low-generation regime typical of RLVR. Theoretically, we construct a shrinkage-based baseline that provably yields lower-variance policy-gradient estimators across algorithms. Our baseline is a drop-in replacement for standard per-prompt mean baselines and requires no additional hyperparameters or computation. Empirically, shrinkage baselines consistently outperform empirical-mean baselines, producing lower-variance gradient updates and improved training stability.
format Preprint
id arxiv_https___arxiv_org_abs_2511_03710
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Shrinking the Variance: Shrinkage Baselines for Reinforcement Learning with Verifiable Rewards
Zeng, Guanning
Zhou, Zhaoyi
Arora, Daman
Zanette, Andrea
Machine Learning
Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a powerful paradigm for post-training large reasoning models (LRMs) using policy-gradient methods such as GRPO. To stabilize training, these methods typically center trajectory rewards by subtracting the empirical mean reward for each prompt. Statistically, this centering acts as a control variate (baseline), reducing the variance of the policy-gradient estimator. In practice, the mean reward is estimated using per-prompt empirical averages computed from the generations for each prompt in a batch. Motivated by Stein's paradox, we propose shrinkage estimators that combine per-prompt and across-prompt means to improve per-prompt mean estimation accuracy, especially in the low-generation regime typical of RLVR. Theoretically, we construct a shrinkage-based baseline that provably yields lower-variance policy-gradient estimators across algorithms. Our baseline is a drop-in replacement for standard per-prompt mean baselines and requires no additional hyperparameters or computation. Empirically, shrinkage baselines consistently outperform empirical-mean baselines, producing lower-variance gradient updates and improved training stability.
title Shrinking the Variance: Shrinkage Baselines for Reinforcement Learning with Verifiable Rewards
topic Machine Learning
url https://arxiv.org/abs/2511.03710