empirical Bayes
2 published articles
Artificial Intelligence
The fraud detector that got better by ignoring what it was told
PREF-Gate separates fraud detection into a label-free context expert and a label-derived evidence expert, then lets a validation gate decide which to deploy. On Amazon and YelpChi, it drops the evidence entirely; only on TFinance does it accept a small mixture. The paper forces the field to confront whether label leakage, not model architecture, drives many published GNN gains.
2026-07-29
Label leakage in GNNs
Your fraud detection model may be cheating. A new paper proves it.
The PREF-Gate paper formalizes a label-provenance contract for relational evidence in graph fraud detection. On two of three datasets, the best model is one that avoids training-label-derived evidence entirely, a result that challenges common assumptions in graph neural network research.
2026-07-26