Multi-criteria Selection of Rehearsal Samples for Continual Learning

作者：

Highlights：

• We present a multi-criteria subset selection strategy to overcome the unstable learning issue of using singular criteria.

• Two novel subset selection strategies are introduced: intra-class cluster variation and classifier loss to replay-based continual learning framework.

• Proposed method achieves new state-of-the-art results and provides an insight into sample selection.

摘要

•We present a multi-criteria subset selection strategy to overcome the unstable learning issue of using singular criteria.•Two novel subset selection strategies are introduced: intra-class cluster variation and classifier loss to replay-based continual learning framework.•Proposed method achieves new state-of-the-art results and provides an insight into sample selection.

论文关键词：Continual Learning,Multiple Criteria,Rehersal Method,Learning to learn

论文评审过程：Received 29 January 2022, Revised 16 June 2022, Accepted 14 July 2022, Available online 20 July 2022, Version of Record 26 July 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108907