Enhanced Few-Shot Class-Incremental Learning via Ensemble Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhu, Mingli, Zhu, Zihao, Chen, Sihong, Chen, Chen, Wu, Baoyuan
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913275582611456
author Zhu, Mingli
Zhu, Zihao
Chen, Sihong
Chen, Chen
Wu, Baoyuan
author_facet Zhu, Mingli
Zhu, Zihao
Chen, Sihong
Chen, Chen
Wu, Baoyuan
contents Few-shot class-incremental learning (FSCIL) aims to continually fit new classes with limited training data, while maintaining the performance of previously learned classes. The main challenges are overfitting the rare new training samples and forgetting old classes. While catastrophic forgetting has been extensively studied, the overfitting problem has attracted less attention in FSCIL. To tackle overfitting challenge, we design a new ensemble model framework cooperated with data augmentation to boost generalization. In this way, the enhanced model works as a library storing abundant features to guarantee fast adaptation to downstream tasks. Specifically, the multi-input multi-output ensemble structure is applied with a spatial-aware data augmentation strategy, aiming at diversifying the feature extractor and alleviating overfitting in incremental sessions. Moreover, self-supervised learning is also integrated to further improve the model generalization. Comprehensive experimental results show that the proposed method can indeed mitigate the overfitting problem in FSCIL, and outperform the state-of-the-art methods.
format Preprint
id arxiv_https___arxiv_org_abs_2401_07208
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Enhanced Few-Shot Class-Incremental Learning via Ensemble Models
Zhu, Mingli
Zhu, Zihao
Chen, Sihong
Chen, Chen
Wu, Baoyuan
Computer Vision and Pattern Recognition
Few-shot class-incremental learning (FSCIL) aims to continually fit new classes with limited training data, while maintaining the performance of previously learned classes. The main challenges are overfitting the rare new training samples and forgetting old classes. While catastrophic forgetting has been extensively studied, the overfitting problem has attracted less attention in FSCIL. To tackle overfitting challenge, we design a new ensemble model framework cooperated with data augmentation to boost generalization. In this way, the enhanced model works as a library storing abundant features to guarantee fast adaptation to downstream tasks. Specifically, the multi-input multi-output ensemble structure is applied with a spatial-aware data augmentation strategy, aiming at diversifying the feature extractor and alleviating overfitting in incremental sessions. Moreover, self-supervised learning is also integrated to further improve the model generalization. Comprehensive experimental results show that the proposed method can indeed mitigate the overfitting problem in FSCIL, and outperform the state-of-the-art methods.
title Enhanced Few-Shot Class-Incremental Learning via Ensemble Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2401.07208