ImpressLearn: Continual Learning via Combined Task Impressions

Dhrupad Bhardwaj, Julia Kempe, Artem M Vysogorets, Angela Teng, Evaristus Ezekwem

Keywords: Catastrophic forgetting, Continual learning, Neural networks, Masking

Abstract:

This work proposes a new method to sequentially train a deep neural network on multiple tasks without suffering catastrophic forgetting, while endowing it with the capability to quickly adapt to unknown tasks. Starting from existing work on network masking (Wortsman et al., 2020), we show that a simple to learn linear combination of a small number of task-specific masks (”impressions”) ona randomly initialized backbone network is sufficient to both retain accuracy on previously learned tasks, as well as achieve high accuracy on new tasks.

In contrast to previous methods, we do not require to generate dedicated masks or contexts for each new task, instead leveraging transfer learning to keep per-task parameter overhead negligible. Our work illustrates the power of linearly combining individual impressions, each of which fares poorly in isolation, to achieve performance comparable to a dedicated mask. Moreover, even repeated impressions from the same task (homogeneous masks), when combined can approach the performance of heterogeneous combinations if sufficiently many impressions are used.

Our approach scales more efficiently than existing methods, requiring orders of magnitude fewer parameters and can function without modification even when task identity is missing. In addition, in the setting where task labels are not given at inference, our algorithm gives an often favorable alternative to the entropy based task-inference methods proposed in (Wortsman et al., 2020). We evaluate our method on a number of well known image classification data sets and architectures

One-sentence Summary: A novel deep learning approach for continual learning that counters catastrophic forgetting using several orders of magnitude fewer parameters than prevailing methods.