Reproduce “Revisiting Class-Incremental Learning with Pre-Trained Models: Generalizability and Adaptivity are All You Need” 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
Da-Wei Zhou, Zi-Wen Cai, Han-Jia Ye, De-Chuan Zhan, Ziwei Liu
APER addresses exemplar-free class-incremental learning by adapting a pretrained model on the first task, then combining its features with those of the original model. It freezes both branches for later tasks and builds classifiers from class-mean features. On ImageNet-A with ViT-B/16-IN21K under B0 Inc10, Adapter APER reaches 64.33% final-stage accuracy versus SimpleCIL’s 49.44%; adaptation occurs only on the first task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.