Reproduce “RASP: Read-All, Sparse-Plasticity Continual Learning in a Stationary-Size LoRA” 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
Wen-Hsuan Liu, Amara Okonkwo, Tom Saliencro, Rohan Desai, Maya Lindqvist
RASP addresses rehearsal-free class-incremental image learning by using a shared adapter that reads all its channels for prediction but updates only selected channels during learning, with consolidation protecting locked channel pairs. On CUB-200, it reaches 81.36% final average accuracy versus SplitLoRA’s 72.88% under the reported 10-session setup, using 0.25M versus 1.53M adapter parameters; on ImageNet-R, SplitLoRA scores higher.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.