Learning label smoothing for text classification

Training with soft labels instead of hard labels can effectively improve the robustness and generalization of deep learning models. Label smoothing often provides uniformly distributed soft labels during the training process, whereas it does not take the semantic difference of labels into account. T...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:PeerJ. Computer science Jg. 10; S. e2005
Hauptverfasser: Ren, Han, Zhao, Yajie, Zhang, Yong, Sun, Wei
Format: Journal Article
Sprache:Englisch
Veröffentlicht: United States PeerJ. Ltd 23.04.2024
PeerJ Inc
Schlagworte:
Algorithms and Analysis of Algorithms
Analysis
Computational Linguistics
Data Mining and Machine Learning
Excessive regularization
Label smoothing
Language processing
Natural language interfaces
Neural network
Neural Networks
Soft label
Text classification
Text Mining
New York
Neural network
Label smoothing
Excessive regularization
Soft label
Text classification
ISSN:2376-5992, 2376-5992
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Training with soft labels instead of hard labels can effectively improve the robustness and generalization of deep learning models. Label smoothing often provides uniformly distributed soft labels during the training process, whereas it does not take the semantic difference of labels into account. This article introduces discrimination-aware label smoothing, an adaptive label smoothing approach that learns appropriate distributions of labels for iterative optimization objectives. In this approach, positive and negative samples are employed to provide experience from both sides, and the performances of regularization and model calibration are improved through an iterative learning method. Experiments on five text classification datasets demonstrate the effectiveness of the proposed method.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:2376-5992
2376-5992
DOI:10.7717/peerj-cs.2005