ZS-IL: Looking Back on Learned Experiences For Zero-Shot Incremental Learning

TL;DR

This paper introduces ZS-IL, a zero-shot incremental learning method that synthesizes past experiences on demand without fixed memory, significantly reducing forgetting in neural networks during sequential task learning.

Contribution

It proposes a novel memory recovery paradigm enabling zero-shot replay of past classes without fixed memory or parallel architectures.

Findings

Outperforms state-of-the-art methods on CIFAR-10 and Tiny-ImageNet.

Effectively mitigates catastrophic forgetting in incremental learning.

Operates without storing actual past data samples.

Abstract

Classical deep neural networks are limited in their ability to learn from emerging streams of training data. When trained sequentially on new or evolving tasks, their performance degrades sharply, making them inappropriate in real-world use cases. Existing methods tackle it by either storing old data samples or only updating a parameter set of DNNs, which, however, demands a large memory budget or spoils the flexibility of models to learn the incremented class distribution. In this paper, we shed light on an on-call transfer set to provide past experiences whenever a new class arises in the data stream. In particular, we propose a Zero-Shot Incremental Learning not only to replay past experiences the model has learned but also to perform this in a zero-shot manner. Towards this end, we introduced a memory recovery paradigm in which we query the network to synthesize past exemplars whenever a new task (class) emerges. Thus, our method needs no fixed-sized memory, besides calls the proposed memory recovery paradigm to provide past exemplars, named a transfer set in order to mitigate catastrophically forgetting the former classes. Moreover, in contrast with recently proposed methods, the suggested paradigm does not desire a parallel architecture since it only relies on the learner network. Compared to the state-of-the-art data techniques without buffering past data samples, ZS-IL demonstrates significantly better performance on the well-known datasets (CIFAR-10, Tiny-ImageNet) in both Task-IL and Class-IL settings.