Reproduce “Unlearning at Speed: Rapid Activation Dampening for Efficient Machine Unlearning in Real-World Face Recognition” 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
Rouqaiah Al-Refai, Leo Scheubel, Philipp Terhörst
Rapid Activation Dampening (RAD) targets a face-recognition identity without retraining by comparing neuron activations for that identity and retained identities, then dampening neurons more responsive to the target. On ArcFace50, RAD-10K scored 0.82 on the Forget Score (FS) in 19.77 seconds, versus LFSSD-E’s 0.97 in 12,954.11 seconds; FS measures relative embedding shifts, not identity erasure, across five requests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.