Class-Incremental Learning via Knowledge Amalgamation

TL;DR

This paper introduces a knowledge amalgamation approach for class-incremental learning that mitigates catastrophic forgetting by learning from multiple heterogeneous teacher models without requiring shared network structures or annotations.

Contribution

It proposes a novel knowledge amalgamation method that enables learning from diverse teacher models in a single-head setting without shared architectures or annotations.

Findings

Outperforms existing methods on incremental learning benchmarks.

Effectively handles heterogeneous teacher models and data representations.

Reduces catastrophic forgetting in continual learning scenarios.

Abstract

Catastrophic forgetting has been a significant problem hindering the deployment of deep learning algorithms in the continual learning setting. Numerous methods have been proposed to address the catastrophic forgetting problem where an agent loses its generalization power of old tasks while learning new tasks. We put forward an alternative strategy to handle the catastrophic forgetting with knowledge amalgamation (CFA), which learns a student network from multiple heterogeneous teacher models specializing in previous tasks and can be applied to current offline methods. The knowledge amalgamation process is carried out in a single-head manner with only a selected number of memorized samples and no annotations. The teachers and students do not need to share the same network structure, allowing heterogeneous tasks to be adapted to a compact or sparse data representation. We compare our method with competitive baselines from different strategies, demonstrating our approach's advantages.