Reproduce “Anchor Behavior, Not Parameters: Rehearsal-Free Continual Learning with a Single Shared Prompt” 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
Duy-Cuong Bui, Trung-Anh Dang, Vincent Nguyen, Duc V. Le, Ngoc-Son Vu
TEA-Prompt adapts a frozen vision transformer for rehearsal-free class-incremental recognition with one shared prompt. Input-dependent relevance weights guide an anchor on prompt-token attention contributions for earlier classes, using class-wise query statistics rather than old images. On 10-task CIFAR-100, it reports 90.59 ± 0.31 average accuracy and 2.68 ± 0.25 forgetting, versus 88.86 and 4.57 for same-size SMoPE in this comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.