DeepCore: A Comprehensive Library for Coreset Selection in Deep Learning

Coreset selection, which aims to select a subset of the most informative training samples, is long-standing learning problem that can benefit many downstream tasks such as data-efficient learning, continual neural architecture search, active etc. However, existing coreset selection methods are not designed for deep may have high complexity and poor generalization performance. In addition, recently proposed evaluated on models, datasets, settings different complexities. To advance research in we contribute comprehensive code library (The available https://github.com/PatrickZH/DeepCore .), namely DeepCore, provide an empirical study popular CIFAR10 ImageNet datasets. Extensive experiments datasets verify that, although various advantages certain experiment settings, random still strong baseline.