Reproduce “Two-Way Is Better Than One: Bidirectional Alignment with Cycle Consistency 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
Hongye Xu, Bartosz Krawczyk
BiCyc learns bidirectional feature maps with cycle consistency during training, transporting old class statistics as representations shift. On scratch-trained TinyImageNet with 10 tasks, it reaches 35.4 ± 0.8% last-task average accuracy and 49.1 ± 1.4% average incremental accuracy, exceeding EFC in this setting. It does not lead on pretrained CUB-200, where it trails EFC.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.