Anomaly Detection Requires Better Representations

TL;DR

This paper discusses how self-supervised representations enhance anomaly detection performance and argues that future progress depends on developing new representation learning techniques.

Contribution

It highlights the current success of self-supervised representations in anomaly detection and emphasizes the need for novel methods to address upcoming challenges.

Findings

Self-supervised representations achieve state-of-the-art results in benchmarks.

Current methods are effective but need improvement for future tasks.

The paper advocates for new representation learning approaches.

Abstract

Anomaly detection seeks to identify unusual phenomena, a central task in science and industry. The task is inherently unsupervised as anomalies are unexpected and unknown during training. Recent advances in self-supervised representation learning have directly driven improvements in anomaly detection. In this position paper, we first explain how self-supervised representations can be easily used to achieve state-of-the-art performance in commonly reported anomaly detection benchmarks. We then argue that tackling the next generation of anomaly detection tasks requires new technical and conceptual improvements in representation learning.