Saved in:
Bibliographic Details
Main Authors: Xie, Zhengpeng, Cao, Jiahang, Wang, Changwei, Yang, Fan, Hutter, Marco, Zhang, Qiang, Zhang, Jianxiong, Xu, Renjing
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.02481
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914053155192832
author Xie, Zhengpeng
Cao, Jiahang
Wang, Changwei
Yang, Fan
Hutter, Marco
Zhang, Qiang
Zhang, Jianxiong
Xu, Renjing
author_facet Xie, Zhengpeng
Cao, Jiahang
Wang, Changwei
Yang, Fan
Hutter, Marco
Zhang, Qiang
Zhang, Jianxiong
Xu, Renjing
contents In this paper, we argue that mutual distillation between reinforcement learning policies serves as an implicit regularization, preventing them from overfitting to irrelevant features. We highlight two separate contributions: (i) Theoretically, for the first time, we prove that enhancing the policy robustness to irrelevant features leads to improved generalization performance. (ii) Empirically, we demonstrate that mutual distillation between policies contributes to such robustness, enabling the spontaneous emergence of invariant representations over pixel inputs. Ultimately, we do not claim to achieve state-of-the-art performance but rather focus on uncovering the underlying principles of generalization and deepening our understanding of its mechanisms.
format Preprint
id arxiv_https___arxiv_org_abs_2501_02481
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Representation Convergence: Mutual Distillation is Secretly a Form of Regularization
Xie, Zhengpeng
Cao, Jiahang
Wang, Changwei
Yang, Fan
Hutter, Marco
Zhang, Qiang
Zhang, Jianxiong
Xu, Renjing
Machine Learning
Artificial Intelligence
In this paper, we argue that mutual distillation between reinforcement learning policies serves as an implicit regularization, preventing them from overfitting to irrelevant features. We highlight two separate contributions: (i) Theoretically, for the first time, we prove that enhancing the policy robustness to irrelevant features leads to improved generalization performance. (ii) Empirically, we demonstrate that mutual distillation between policies contributes to such robustness, enabling the spontaneous emergence of invariant representations over pixel inputs. Ultimately, we do not claim to achieve state-of-the-art performance but rather focus on uncovering the underlying principles of generalization and deepening our understanding of its mechanisms.
title Representation Convergence: Mutual Distillation is Secretly a Form of Regularization
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2501.02481