Anomaly Detection With Partitioning Overfitting Autoencoder Ensembles

TL;DR

This paper introduces POTATOES, a novel unsupervised outlier detection method that leverages overfitting autoencoder ensembles with data partitioning to enhance detection accuracy without tuning regularization.

Contribution

The paper presents a new ensemble-based approach that improves autoencoder outlier detection by partitioning data and overfitting autoencoders, eliminating the need for regularization tuning.

Findings

Significant improvement in UOD performance on various datasets.

Effectiveness depends on the density of inliers.

Method is applicable to different autoencoders and datasets.

Abstract

In this paper, we propose POTATOES (Partitioning OverfiTting AuTOencoder EnSemble), a new method for unsupervised outlier detection (UOD). More precisely, given any autoencoder for UOD, this technique can be used to improve its accuracy while at the same time removing the burden of tuning its regularization. The idea is to not regularize at all, but to rather randomly partition the data into sufficiently many equally sized parts, overfit each part with its own autoencoder, and to use the maximum over all autoencoder reconstruction errors as the anomaly score. We apply our model to various realistic datasets and show that if the set of inliers is dense enough, our method indeed improves the UOD performance of a given autoencoder significantly. For reproducibility, the code is made available on github so the reader can recreate the results in this paper as well as apply the method to other autoencoders and datasets.