B-STaR: Monitoring and Balancing Exploration and Exploitation in Self-Taught Reasoners

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zeng, Weihao, Huang, Yuzhen, Zhao, Lulu, Wang, Yijun, Shan, Zifei, He, Junxian
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913717563686912
author Zeng, Weihao
Huang, Yuzhen
Zhao, Lulu
Wang, Yijun
Shan, Zifei
He, Junxian
author_facet Zeng, Weihao
Huang, Yuzhen
Zhao, Lulu
Wang, Yijun
Shan, Zifei
He, Junxian
contents In the absence of extensive human-annotated data for complex reasoning tasks, self-improvement -- where models are trained on their own outputs -- has emerged as a primary method for enhancing performance. However, the critical factors underlying the mechanism of these iterative self-improving methods remain poorly understood, such as under what conditions self-improvement is effective, and what are the bottlenecks in the current iterations. In this work, we identify and propose methods to monitor two pivotal factors in this iterative process: (1) the model's ability to generate sufficiently diverse responses (exploration); and (2) the effectiveness of external rewards in distinguishing high-quality candidates from lower-quality ones (exploitation). Using mathematical reasoning as a case study, we begin with a quantitative analysis to track the dynamics of exploration and exploitation, discovering that a model's exploratory capabilities rapidly deteriorate over iterations, and the effectiveness of exploiting external rewards diminishes as well. Motivated by these findings, we introduce B-STaR, a Self-Taught Reasoning framework that autonomously adjusts configurations across iterations to Balance exploration and exploitation, thereby optimizing the self-improving effectiveness based on the current policy model and available rewards. Our experiments on mathematical reasoning, coding, and commonsense reasoning demonstrate that B-STaR not only enhances the model's exploratory capabilities throughout training but also achieves a more effective balance between exploration and exploitation, leading to superior performance.
format Preprint
id arxiv_https___arxiv_org_abs_2412_17256
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle B-STaR: Monitoring and Balancing Exploration and Exploitation in Self-Taught Reasoners
Zeng, Weihao
Huang, Yuzhen
Zhao, Lulu
Wang, Yijun
Shan, Zifei
He, Junxian
Artificial Intelligence
Computation and Language
Machine Learning
In the absence of extensive human-annotated data for complex reasoning tasks, self-improvement -- where models are trained on their own outputs -- has emerged as a primary method for enhancing performance. However, the critical factors underlying the mechanism of these iterative self-improving methods remain poorly understood, such as under what conditions self-improvement is effective, and what are the bottlenecks in the current iterations. In this work, we identify and propose methods to monitor two pivotal factors in this iterative process: (1) the model's ability to generate sufficiently diverse responses (exploration); and (2) the effectiveness of external rewards in distinguishing high-quality candidates from lower-quality ones (exploitation). Using mathematical reasoning as a case study, we begin with a quantitative analysis to track the dynamics of exploration and exploitation, discovering that a model's exploratory capabilities rapidly deteriorate over iterations, and the effectiveness of exploiting external rewards diminishes as well. Motivated by these findings, we introduce B-STaR, a Self-Taught Reasoning framework that autonomously adjusts configurations across iterations to Balance exploration and exploitation, thereby optimizing the self-improving effectiveness based on the current policy model and available rewards. Our experiments on mathematical reasoning, coding, and commonsense reasoning demonstrate that B-STaR not only enhances the model's exploratory capabilities throughout training but also achieves a more effective balance between exploration and exploitation, leading to superior performance.
title B-STaR: Monitoring and Balancing Exploration and Exploitation in Self-Taught Reasoners
topic Artificial Intelligence
Computation and Language
Machine Learning
url https://arxiv.org/abs/2412.17256