Reproduce “Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection” 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
Amey Pasarkar, Adji Bousso Dieng
The authors propose Vendi Anomaly Score (VAS), which ranks unlabeled samples by how much removing each one changes similarity-based dataset diversity. A first-order spectral approximation enables batched scoring and selects its spectrum order automatically. In controlled, class-disjoint ImageNet experiments, VAS leads tested methods across contamination-rate and dataset-size sweeps, but this does not establish performance on all real-world anomalies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.