Reproduce “Auditing Representation-Space Fraud Detection: A Bayesian IRL Framework for Calibrated Uncertainty and Its Limits” 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

Shivam Swarup, Divya Prakash Shrivastava, Rakesh Thakur

The authors introduce VEIL, an auditing framework that tests whether recommender-system fraud detectors retain evidence of shilling attacks in frozen user-profile embeddings. It uses Bayesian pairwise preference models to measure ambiguity and confidence. On MovieLens-1M donor-mimicry, VEIL’s mean AUROC is 0.494 ± 0.061, while raw-profile Jaccard reaches 0.974; this shows a representation-specific failure, not universal detectability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.