Reproduce “Adapting prior-data fitted networks for tabular 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
Maximilian Bershtman, Niv Cohen
FOCUS adapts TabPFN-3 representations for unsupervised tabular anomaly detection by fine-tuning embeddings without anomaly labels. Frozen-feature ZEN scores rows by distances to neighbors in those representations and downweights isolated reference rows. On contaminated ADBench, FOCUS achieves mean AUROC of 80.0, versus 76.4 for Isolation Forest; ZEN scores 79.7. These are averages across 47 datasets, not guarantees across datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.