Improving Value-based Process Verifier via Low-Cost Variance Reduction

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sun, Zetian, Li, Dongfang, Hu, Baotian, Zhang, Min
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912850775113728
author Sun, Zetian
Li, Dongfang
Hu, Baotian
Zhang, Min
author_facet Sun, Zetian
Li, Dongfang
Hu, Baotian
Zhang, Min
contents Large language models (LLMs) have achieved remarkable success in a wide range of tasks. However, their reasoning capabilities, particularly in complex domains like mathematics, remain a significant challenge. Value-based process verifiers, which estimate the probability of a partial reasoning chain leading to a correct solution, are a promising approach for improving reasoning. Nevertheless, their effectiveness is often hindered by estimation error in their training annotations, a consequence of the limited number of Monte Carlo (MC) samples feasible due to the high cost of LLM inference. In this paper, we identify that the estimation error primarily arises from high variance rather than bias, and the MC estimator is a Minimum Variance Unbiased Estimator (MVUE). To address the problem, we propose the \textsc{Com}pound \textsc{M}onte \textsc{C}arlo \textsc{S}ampling (ComMCS) method, which constructs an unbiased estimator by linearly combining the MC estimators from the current and subsequent steps. Theoretically, we show that our method leads to a predictable reduction in variance, while maintaining an unbiased estimation without additional LLM inference cost. We also perform empirical experiments on the MATH-500 and GSM8K benchmarks to demonstrate the effectiveness of our method. Notably, ComMCS outperforms regression-based optimization method by 2.8 points, the non-variance-reduced baseline by 2.2 points on MATH-500 on Best-of-32 sampling experiment.
format Preprint
id arxiv_https___arxiv_org_abs_2508_10539
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Improving Value-based Process Verifier via Low-Cost Variance Reduction
Sun, Zetian
Li, Dongfang
Hu, Baotian
Zhang, Min
Artificial Intelligence
Computation and Language
Large language models (LLMs) have achieved remarkable success in a wide range of tasks. However, their reasoning capabilities, particularly in complex domains like mathematics, remain a significant challenge. Value-based process verifiers, which estimate the probability of a partial reasoning chain leading to a correct solution, are a promising approach for improving reasoning. Nevertheless, their effectiveness is often hindered by estimation error in their training annotations, a consequence of the limited number of Monte Carlo (MC) samples feasible due to the high cost of LLM inference. In this paper, we identify that the estimation error primarily arises from high variance rather than bias, and the MC estimator is a Minimum Variance Unbiased Estimator (MVUE). To address the problem, we propose the \textsc{Com}pound \textsc{M}onte \textsc{C}arlo \textsc{S}ampling (ComMCS) method, which constructs an unbiased estimator by linearly combining the MC estimators from the current and subsequent steps. Theoretically, we show that our method leads to a predictable reduction in variance, while maintaining an unbiased estimation without additional LLM inference cost. We also perform empirical experiments on the MATH-500 and GSM8K benchmarks to demonstrate the effectiveness of our method. Notably, ComMCS outperforms regression-based optimization method by 2.8 points, the non-variance-reduced baseline by 2.2 points on MATH-500 on Best-of-32 sampling experiment.
title Improving Value-based Process Verifier via Low-Cost Variance Reduction
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2508.10539