Gradient-matching coresets for continual learning

Abstract

We devise a coreset selection method based on the idea of gradient matching: the gradients induced by the coreset should match, as closely as possible, those induced by the original training dataset. We evaluate the method in the context of continual learning, where it can be used to curate a rehearsal memory. Our method performs strong competitors such as reservoir sampling across a range of memory sizes. Related workA number of methods for efficientily managing memory in continual learning have been proposed over time (e.g., see Borsos et al. [2020], Aljundi et al. [2019a,b]). Due to space constraints we will discuss only the ones which are closely related to our work.