Abstract:Novelty detection, a widely studied problem in machine learning, is the problem of detecting a novel class of data that has not been previously observed. A common setting for novelty detection is inductive whereby only examples of the negative class are available during training time. Transductive novelty detection on the other hand has only witnessed a recent surge in interest, it not only makes use of the negative class during training but also incorporates the (unlabeled) test set to detect novel examples. Several studies have emerged under the transductive setting umbrella that have demonstrated its advantage over its inductive counterpart. Depending on the assumptions about the data, these methods go by different names (e.g. transductive novelty detection, semi-supervised novelty detection, positive-unlabeled learning, out-of-distribution detection). With the use of generative adversarial networks (GAN), a segment of those studies have adopted a transductive setup in order to learn how to generate examples of the novel class. In this study, we propose TransductGAN, a transductive generative adversarial network that attempts to learn how to generate image examples from both the novel and negative classes by using a mixture of two Gaussians in the latent space. It achieves that by incorporating an adversarial autoencoder with a GAN network, the ability to generate examples of novel data points offers not only a visual representation of novelties, but also overcomes the hurdle faced by many inductive methods of how to tune the model hyperparameters at the decision rule level. Our model has shown superior performance over state-of-the-art inductive and transductive methods. Our study is fully reproducible with the code available publicly.

Generative Adversarial Network Based Novelty Detection Usingminimized Reconstruction Error

TransductGAN: a Transductive Adversarial Model for Novelty Detection

Novelty Detection for Multimode Process Using GANs with Learning Disentangled Representation

on the effectiveness of generative adversarial network on anomaly detection

Adversarially Learned Anomaly Detection

Generative Models for Novelty Detection: Applications in abnormal event and situational change detection from data series

Detecting Anomalies using Generative Adversarial Networks on Images

Video Anomaly Detection using GAN

MIM-Based Generative Adversarial Networks and Its Application on Anomaly Detection

MIM-Based Generative Adversarial Networks and Its Application on Anomaly Detection.

Are generative deep models for novelty detection truly better?

MDGAN: Boosting Anomaly Detection Using \\Multi-Discriminator Generative Adversarial Networks

Unsupervised Anomaly Detection Via Generative Adversarial Networks: Poster Abstract

Beyond the Known: Adversarial Autoencoders in Novelty Detection

Adversarially Learned One-Class Classifier for Novelty Detection

Two-stage anomaly detection for positive samples and small samples based on generative adversarial networks

GAN Ensemble for Anomaly Detection

Attention-guided generator with dual discriminator GAN for real-time video anomaly detection

GANomaly: Semi-supervised Anomaly Detection via Adversarial Training

Generative adversarial networks

Efficient GAN-based method for cyber-intrusion detection