TracInAD: Measuring Influence for Anomaly Detection

TL;DR

We address anomaly detection on tabular data by leveraging TracIn influence scores to identify anomalies, using a Variational Autoencoder as the testbed. The method, TracInAD, computes an anomaly score from the average influence of training samples on test points at training checkpoints, enabling a label-agnostic detection signal. Empirical results on four tabular benchmarks show competitive or superior performance to state-of-the-art methods, with Thyroid often strongest, and demonstrate the approach's generality by applying it to Deep SVDD as well. The work highlights the potential of influence-based explanations as practical anomaly-score augmentations for deep unsupervised detectors.

Abstract

As with many other tasks, neural networks prove very effective for anomaly detection purposes. However, very few deep-learning models are suited for detecting anomalies on tabular datasets. This paper proposes a novel methodology to flag anomalies based on TracIn, an influence measure initially introduced for explicability purposes. The proposed methods can serve to augment any unsupervised deep anomaly detection method. We test our approach using Variational Autoencoders and show that the average influence of a subsample of training points on a test point can serve as a proxy for abnormality. Our model proves to be competitive in comparison with state-of-the-art approaches: it achieves comparable or better performance in terms of detection accuracy on medical and cyber-security tabular benchmark data.