Multi-granularity episodic contrastive learning for few-shot learning

作者：

Highlights：

• We develop a novel multi-granularity episodic contrastive learning method and propose two granularities of contrastive loss to learn category-independent discriminative patterns.

• Our method successfully combines the advantages of pre-training and meta-learning, making adaptation to novel classes much easier.

• Extensive experiments are conducted on multiple popular benchmarks for FSL to illustrate the effectiveness and superiority of our method.

摘要

•We develop a novel multi-granularity episodic contrastive learning method and propose two granularities of contrastive loss to learn category-independent discriminative patterns.•Our method successfully combines the advantages of pre-training and meta-learning, making adaptation to novel classes much easier.•Extensive experiments are conducted on multiple popular benchmarks for FSL to illustrate the effectiveness and superiority of our method.

论文关键词：Multi-granularity computing,Episodic contrastive learning,Few-shot learning,Deep learning

论文评审过程：Received 16 October 2021, Revised 13 April 2022, Accepted 28 May 2022, Available online 31 May 2022, Version of Record 21 June 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108820