Representation Learning in Anomaly Detection: Successes, Limits and a Grand Challenge
This perspective identifies critical limitations in anomaly detection for researchers and practitioners, highlighting unsolved problems in scientific and image domains.
The paper argues that the dominant anomaly detection paradigm faces fundamental scaling limits due to a no free lunch principle, but can succeed with strong task priors in industrial settings, while posing scientific discovery and ImageNet anomaly detection as grand challenges requiring new tools.
In this perspective paper, we argue that the dominant paradigm in anomaly detection cannot scale indefinitely and will eventually hit fundamental limits. This is due to the a no free lunch principle for anomaly detection. These limitations can be overcome when there are strong tasks priors, as is the case for many industrial tasks. When such priors do not exists, the task is much harder for anomaly detection. We pose two such tasks as grand challenges for anomaly detection: i) scientific discovery by anomaly detection ii) a "mini-grand" challenge of detecting the most anomalous image in the ImageNet dataset. We believe new anomaly detection tools and ideas would need to be developed to overcome these challenges.