Reproduce “SPARS: A Stability-Plasticity-Aware Region Selection Framework for 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
Teif Aldaajani, Huda Hakami, Maha A. Thafar
SPARS protects important weights during continual learning by freezing top-ranked parameters, softly regularizing neighboring convolution-kernel weights, and leaving others free. It estimates importance with labeled Fisher information or unlabeled confidence sensitivity. On TinyImageNet class-incremental learning, SPARS-FIM reaches 36.13% average incremental accuracy, compared with EDD's 30.11%, but does not lead on CIFAR-10; these results cover two image benchmarks with ResNet-18.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.