English

Few-Shot Incremental Learning with Continually Evolved Classifiers

Computer Vision and Pattern Recognition 2021-04-08 v1 Artificial Intelligence Machine Learning

Abstract

Few-shot class-incremental learning (FSCIL) aims to design machine learning algorithms that can continually learn new concepts from a few data points, without forgetting knowledge of old classes. The difficulty lies in that limited data from new classes not only lead to significant overfitting issues but also exacerbate the notorious catastrophic forgetting problems. Moreover, as training data come in sequence in FSCIL, the learned classifier can only provide discriminative information in individual sessions, while FSCIL requires all classes to be involved for evaluation. In this paper, we address the FSCIL problem from two aspects. First, we adopt a simple but effective decoupled learning strategy of representations and classifiers that only the classifiers are updated in each incremental session, which avoids knowledge forgetting in the representations. By doing so, we demonstrate that a pre-trained backbone plus a non-parametric class mean classifier can beat state-of-the-art methods. Second, to make the classifiers learned on individual sessions applicable to all classes, we propose a Continually Evolved Classifier (CEC) that employs a graph model to propagate context information between classifiers for adaptation. To enable the learning of CEC, we design a pseudo incremental learning paradigm that episodically constructs a pseudo incremental learning task to optimize the graph parameters by sampling data from the base dataset. Experiments on three popular benchmark datasets, including CIFAR100, miniImageNet, and Caltech-USCD Birds-200-2011 (CUB200), show that our method significantly outperforms the baselines and sets new state-of-the-art results with remarkable advantages.

Keywords

Cite

@article{arxiv.2104.03047,
  title  = {Few-Shot Incremental Learning with Continually Evolved Classifiers},
  author = {Chi Zhang and Nan Song and Guosheng Lin and Yun Zheng and Pan Pan and Yinghui Xu},
  journal= {arXiv preprint arXiv:2104.03047},
  year   = {2021}
}

Comments

Accpeted to CVPR2021