Reproduce “DLER: Dual-Logit Experience Replay for Online Continual 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

Huiwei Lin, Shanshan Feng, Kong Aik Lee

DLER addresses class-incremental learning by treating incoming and replayed examples differently. It uses cosine logits for incoming samples and adds dot-product supervision for replayed samples, alongside supervised contrastive learning and advantage-based memory selection. On Split TinyImageNet with a buffer of 400 examples, it reports 11.9 ± 0.3% average accuracy, versus 9.7 ± 0.4% for UER; these means are not a significance test.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.