Reproduce “Adversarial Pseudo-replay for Exemplar-free Class-incremental Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hiroto Honda
APR addresses exemplar-free class-incremental learning by generating pseudo-replay online: it steers current-task image features toward saved old-class prototypes and uses the resulting images for distillation, without storing replay samples. On cold-start ImageNet-Subset with 10 tasks and a nearest-class-mean classifier, it reports average incremental/final accuracy of 73.00/60.99, versus ADC’s 69.98/55.83; training takes 31.0 versus 13.5 hours.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.