Memorizing Complementation Network for Few-Shot Class-Incremental Learning

TL;DR

This paper introduces MCNet, a method for few-shot class-incremental learning that ensembles multiple models to better retain old knowledge and adapt to new classes with limited samples, addressing catastrophic forgetting.

Contribution

The paper proposes a novel ensemble approach called MCNet and a Prototype Smoothing Hard-mining Triplet loss for improved few-shot class-incremental learning.

Findings

Outperforms existing methods on CIFAR100, miniImageNet, and CUB200 datasets.

Effectively balances knowledge retention and adaptation with limited data.

Reduces catastrophic forgetting in incremental learning scenarios.

Abstract

Few-shot Class-Incremental Learning (FSCIL) aims at learning new concepts continually with only a few samples, which is prone to suffer the catastrophic forgetting and overfitting problems. The inaccessibility of old classes and the scarcity of the novel samples make it formidable to realize the trade-off between retaining old knowledge and learning novel concepts. Inspired by that different models memorize different knowledge when learning novel concepts, we propose a Memorizing Complementation Network (MCNet) to ensemble multiple models that complements the different memorized knowledge with each other in novel tasks. Additionally, to update the model with few novel samples, we develop a Prototype Smoothing Hard-mining Triplet (PSHT) loss to push the novel samples away from not only each other in current task but also the old distribution. Extensive experiments on three benchmark datasets, e.g., CIFAR100, miniImageNet and CUB200, have demonstrated the superiority of our proposed method.