OAAE: Adversarial Autoencoders for Novelty Detection in Multi-modal Normality Case via Orthogonalized Latent Space

TL;DR

This paper introduces OAAE, an adversarial autoencoder that uses orthogonalized latent space to improve novelty detection in multi-modal normality cases, outperforming existing GAN-based methods.

Contribution

It proposes a novel orthogonal low-rank embedding in the latent space to better disentangle features for improved novelty detection in complex multi-modal data.

Findings

Outperforms state-of-the-art GAN-based novelty detection algorithms

Effective disentanglement of features in multi-modal normality cases

Improved detection accuracy demonstrated through experiments

Abstract

Novelty detection using deep generative models such as autoencoder, generative adversarial networks mostly takes image reconstruction error as novelty score function. However, image data, high dimensional as it is, contains a lot of different features other than class information which makes models hard to detect novelty data. The problem gets harder in multi-modal normality case. To address this challenge, we propose a new way of measuring novelty score in multi-modal normality cases using orthogonalized latent space. Specifically, we employ orthogonal low-rank embedding in the latent space to disentangle the features in the latent space using mutual class information. With the orthogonalized latent space, novelty score is defined by the change of each latent vector. Proposed algorithm was compared to state-of-the-art novelty detection algorithms using GAN such as RaPP and OCGAN, and experimental results show that ours outperforms those algorithms.