MaxSup: Overcoming Representation Collapse in Label Smoothing

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhou, Yuxuan, Li, Heng, Cheng, Zhi-Qi, Yan, Xudong, Dong, Yifei, Fritz, Mario, Keuper, Margret
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914306964062208
author Zhou, Yuxuan
Li, Heng
Cheng, Zhi-Qi
Yan, Xudong
Dong, Yifei
Fritz, Mario
Keuper, Margret
author_facet Zhou, Yuxuan
Li, Heng
Cheng, Zhi-Qi
Yan, Xudong
Dong, Yifei
Fritz, Mario
Keuper, Margret
contents Label Smoothing (LS) is widely adopted to reduce overconfidence in neural network predictions and improve generalization. Despite these benefits, recent studies reveal two critical issues with LS. First, LS induces overconfidence in misclassified samples. Second, it compacts feature representations into overly tight clusters, diluting intra-class diversity, although the precise cause of this phenomenon remained elusive. In this paper, we analytically decompose the LS-induced loss, exposing two key terms: (i) a regularization term that dampens overconfidence only when the prediction is correct, and (ii) an error-amplification term that arises under misclassifications. This latter term compels the network to reinforce incorrect predictions with undue certainty, exacerbating representation collapse. To address these shortcomings, we propose Max Suppression (MaxSup), which applies uniform regularization to both correct and incorrect predictions by penalizing the top-1 logit rather than the ground-truth logit. Through extensive feature-space analyses, we show that MaxSup restores intra-class variation and sharpens inter-class boundaries. Experiments on large-scale image classification and multiple downstream tasks confirm that MaxSup is a more robust alternative to LS. Code is available at: https://github.com/ZhouYuxuanYX/Maximum-Suppression-Regularization
format Preprint
id arxiv_https___arxiv_org_abs_2502_15798
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MaxSup: Overcoming Representation Collapse in Label Smoothing
Zhou, Yuxuan
Li, Heng
Cheng, Zhi-Qi
Yan, Xudong
Dong, Yifei
Fritz, Mario
Keuper, Margret
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Label Smoothing (LS) is widely adopted to reduce overconfidence in neural network predictions and improve generalization. Despite these benefits, recent studies reveal two critical issues with LS. First, LS induces overconfidence in misclassified samples. Second, it compacts feature representations into overly tight clusters, diluting intra-class diversity, although the precise cause of this phenomenon remained elusive. In this paper, we analytically decompose the LS-induced loss, exposing two key terms: (i) a regularization term that dampens overconfidence only when the prediction is correct, and (ii) an error-amplification term that arises under misclassifications. This latter term compels the network to reinforce incorrect predictions with undue certainty, exacerbating representation collapse. To address these shortcomings, we propose Max Suppression (MaxSup), which applies uniform regularization to both correct and incorrect predictions by penalizing the top-1 logit rather than the ground-truth logit. Through extensive feature-space analyses, we show that MaxSup restores intra-class variation and sharpens inter-class boundaries. Experiments on large-scale image classification and multiple downstream tasks confirm that MaxSup is a more robust alternative to LS. Code is available at: https://github.com/ZhouYuxuanYX/Maximum-Suppression-Regularization
title MaxSup: Overcoming Representation Collapse in Label Smoothing
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2502.15798