Continual Learning with Informative Samples: An Empirical Evaluation of Coreset Strategies

Elif Ceren Gok Yildirim, Murat Onur Yildirim, Joaquin Vanschoren

Keywords: continual learning, catastrophic forgetting, coreset selection, data-centric ai

TL;DR: We explore how using only the most important data samples, rather than all available data, can improve continual learning.

Abstract: Continual Learning (CL) methods usually learn from all the available data. However, this is not the case in human cognition which efficiently focuses on key experiences while disregarding the redundant information. Similarly, not all data points in a dataset have equal potential; some can be more informative than others. Especially in CL, such redundant or low-quality data can be detrimental for learning efficiency and exacerbate catastrophic forgetting. Drawing inspiration from this, we explore the potential of learning from important samples and present an empirical study for evaluating coreset selection techniques in the context of CL to stimulate research in this unexplored area. We train various continual learners on progressively larger subsets of selected samples, analyzing the learning-forgetting dynamics and uncovering the mechanisms that enhance the stability-plasticity tradeoff. We present several significant observations: Learning from selectively chosen samples (i) enhances incremental accuracy, (ii) improves knowledge retention of previous tasks, and (iii) continually refines learned representations. This analysis contributes to a deeper understanding of data-selective learning strategies in CL scenarios. The code is available at https://github.com/ElifCerenGokYildirim/Coreset-CL

Submission Number: 4