Saved in:
Bibliographic Details
Main Authors: Zhang, Qingyang, Bian, Yatao, Kong, Xinke, Zhao, Peilin, Zhang, Changqing
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.10894
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916438548152320
author Zhang, Qingyang
Bian, Yatao
Kong, Xinke
Zhao, Peilin
Zhang, Changqing
author_facet Zhang, Qingyang
Bian, Yatao
Kong, Xinke
Zhao, Peilin
Zhang, Changqing
contents Machine learning models must continuously self-adjust themselves for novel data distribution in the open world. As the predominant principle, entropy minimization (EM) has been proven to be a simple yet effective cornerstone in existing test-time adaption (TTA) methods. While unfortunately its fatal limitation (i.e., overconfidence) tends to result in model collapse. For this issue, we propose to Conservatively Minimize the Entropy (COME), which is a simple drop-in replacement of traditional EM to elegantly address the limitation. In essence, COME explicitly models the uncertainty by characterizing a Dirichlet prior distribution over model predictions during TTA. By doing so, COME naturally regularizes the model to favor conservative confidence on unreliable samples. Theoretically, we provide a preliminary analysis to reveal the ability of COME in enhancing the optimization stability by introducing a data-adaptive lower bound on the entropy. Empirically, our method achieves state-of-the-art performance on commonly used benchmarks, showing significant improvements in terms of classification accuracy and uncertainty estimation under various settings including standard, life-long and open-world TTA, i.e., up to $34.5\%$ improvement on accuracy and $15.1\%$ on false positive rate.
format Preprint
id arxiv_https___arxiv_org_abs_2410_10894
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle COME: Test-time adaption by Conservatively Minimizing Entropy
Zhang, Qingyang
Bian, Yatao
Kong, Xinke
Zhao, Peilin
Zhang, Changqing
Machine Learning
Machine learning models must continuously self-adjust themselves for novel data distribution in the open world. As the predominant principle, entropy minimization (EM) has been proven to be a simple yet effective cornerstone in existing test-time adaption (TTA) methods. While unfortunately its fatal limitation (i.e., overconfidence) tends to result in model collapse. For this issue, we propose to Conservatively Minimize the Entropy (COME), which is a simple drop-in replacement of traditional EM to elegantly address the limitation. In essence, COME explicitly models the uncertainty by characterizing a Dirichlet prior distribution over model predictions during TTA. By doing so, COME naturally regularizes the model to favor conservative confidence on unreliable samples. Theoretically, we provide a preliminary analysis to reveal the ability of COME in enhancing the optimization stability by introducing a data-adaptive lower bound on the entropy. Empirically, our method achieves state-of-the-art performance on commonly used benchmarks, showing significant improvements in terms of classification accuracy and uncertainty estimation under various settings including standard, life-long and open-world TTA, i.e., up to $34.5\%$ improvement on accuracy and $15.1\%$ on false positive rate.
title COME: Test-time adaption by Conservatively Minimizing Entropy
topic Machine Learning
url https://arxiv.org/abs/2410.10894