Reproduce “Group and Exclusive Sparse Regularization-based Continual Learning of CNNs” 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
Basile Tousside, Janis Mohr, Jörg Frochte
GESCL addresses sequential image classification with a fixed-capacity CNN after earlier task data becomes unavailable. It protects filters important to past tasks and sparsifies and reinitializes filters judged unimportant for reuse. Across three image benchmarks, it generally retains first-task accuracy, with exceptions; on CIFAR-10/100, removing stability drops average accuracy from 74.5% to 41.9% and raises forgetting from 0.0008 to 0.44.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.