Hugo Lyons Keenan

The paper proposes reframing mechanistic anomaly detection (MAD) as a functional attribution problem, using influence functions to measure how much a model's output depends on specific input samples, achieving state-of-the-art results across various anomaly types and modalities.