How Relevant is Selective Memory Population in Lifelong Language Learning?

TL;DR

This paper investigates the impact of sampling strategies for memory population in lifelong language learning, finding that uniform random sampling performs well, especially with limited memory, aligning with computer vision findings.

Contribution

It is the first to analyze the effect of sampling strategies on memory population in lifelong language learning tasks.

Findings

Uniform random sampling yields high performance with low memory.

Sampling strategy choice has significant impact on lifelong learning outcomes.

Results align with findings in computer vision studies.

Abstract

Lifelong language learning seeks to have models continuously learn multiple tasks in a sequential order without suffering from catastrophic forgetting. State-of-the-art approaches rely on sparse experience replay as the primary approach to prevent forgetting. Experience replay usually adopts sampling methods for the memory population; however, the effect of the chosen sampling strategy on model performance has not yet been studied. In this paper, we investigate how relevant the selective memory population is in the lifelong learning process of text classification and question-answering tasks. We found that methods that randomly store a uniform number of samples from the entire data stream lead to high performances, especially for low memory size, which is consistent with computer vision studies.