Generative Probabilistic Novelty Detection with Adversarial Autoencoders

By Stanislav Pidhorskyi, Ranya Almohsen, D. Adjeroh, Gianfranco Doretto. 2018

349 citations recorded by the source corpus.

Novelty detection is the problem of identifying whether a new data point is considered to be an inlier or an outlier. We assume that training data is available to describe only the inlier distribution. Recent approaches primarily leverage deep encoder-decoder network architectures to compute a reconstruction error that is used to either compute a novelty score or to train a one-class classifier. While we too leverage a novel network of that kind, we take a probabilistic approach and effectively compute how likely it is that a sample was generated by the inlier distribution. We achieve this with two main contributions. First, we make the computation of the novelty probability feasible because we linearize the parameterized manifold capturing the underlying structure of the inlier distribution, and show how the probability factorizes and can be computed with respect to local coordinates of the manifold tangent space. Second, we improve the training of the autoencoder network. An extensive set of results show that the approach achieves state-of-the-art performance on several benchmark datasets.

Subfields: Artificial Intelligence, Epidemiology, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Bing Liu
2. Ziwei Liu
3. Yixuan Li
4. Bin Liu
5. Mohit Bansal
6. Yoshua Bengio
7. Jianfeng Gao
8. Fahad Shahbaz Khan

In Epidemiology

1. Yixuan Li
2. Bin Liu
3. Bing Liu
4. Mengyu Wang
5. Wenpeng Hu
6. Vishal M. Patel
7. Poojan Oza
8. Bernhard Schölkopf

In Molecular Biology

1. Mengyu Wang
2. Wenpeng Hu
3. Vishal M. Patel
4. Bo Li
5. Bing Xiang
6. Poojan Oza
7. Pramuditha Perera
8. Yixuan Li

Citations in the corpus

• Towards Open World Object Detection (2021, background): “In similar spirit, several dedicated approaches target on detecting the out of distribution samples [30] or novelties [48].”
• Out-of-Domain Detection for Natural Language Understanding in Dialog Systems (2019, background): “Popular approaches include modeling the probability density [8], [25], computing reconstruction losses [7], [26],”
• C2AE: Class Conditioned Auto-Encoder for Open-Set Recognition (2019, background): “There are some related problems such as anomaly detection [31], [32], [36] and novelty detection [34], [39], [33], [43] etc.”
• OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, methodology): “Density-based methods model the in-distribution with probabilistic models, which also achieve good performance and are easier for theoretical analysis [15, 16, 17].”
• OCGAN: One-Class Novelty Detection Using GANs With Constrained Latent Representations (2019, methodology): “In [18], first, a GAN is used to obtain a latent representation.”
• Multiple Class Novelty Detection Under Data Distribution Shift (2020, methodology): “Many problems such as countering adversarial/trojan/poison attacks [26,25,8,47], detecting novel categories [42,40,7,37,35,39] and out-of-distribution samples [13,24,10,54] etc. tackle different aspects of robustness of CNNs.”
• Generalized Out-of-Distribution Detection: A Survey (2021, positive): “4.2 can be directly adapted to OOD detection by unifying the ID data as a whole (Zong et al., 2018; Abati et al., 2019; Pidhorskyi et al., 2018; Deecke et al., 2018; Sabokrou et al., 2018).”
• PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization (2020, neutral): “[16], [18], intermediate activations [19] or a discriminator [17], [20] can help to better recognize anomalous images.”
• OpenGAN: Open-Set Recognition via Open Data Generation (2021, neutral): “To recognize open-set examples, they resort to post-hoc functions like density estimation [59, 57], uncertainty modeling [17, 29], and input image reconstruction [39, 20, 14, 50].”
• Towards Total Recall in Industrial Anomaly Detection (2021, methodology): “Other unsupervised representation learning methods can similarly be utilised, such as via GANs [13], learning to predict predefined geometric transformations [20] or via normalizing flows [42].”
• CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (2020, background): “Most recent works can be grouped into four categories: (a) density-based [66, 43, 7, 12, 52, 57, 17], (b) reconstruction-based [55, 68, 10, 50, 48, 8], (c) one-class classifier [56, 53], and (d) self-supervised [15, 26, 1] methods.”
• Is Out-of-Distribution Detection Learnable? (2022, methodology): “A rich line of algorithms have been developed to empirically address the OOD detection problem [6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20].”
• VT-ADL: A Vision Transformer Network for Image Anomaly Detection and Localization (2021, background): “Network architectures mostly consist of various configuration of autoencoders [12], [13], [14], [15], [16] or Generative Adversarial Network (GAN) [17], [18].”
• Detecting semantic anomalies (2019, methodology): “We discourage the recently-adopted practice of treating one category as in-distribution and many other categories as out-distributions (as in [20, 21], for example).”
• CMG: A Class-Mixed Generation Approach to Out-of-Distribution Detection (2022, methodology): “In some other methods, generative models such as auto-encoders [59, 43] and generative adversarial networks (GAN) are used to reconstruct the training data [8, 42].”
• HRN: A Holistic Approach to One Class Learning (2020, background): “More recently, deep learning models have been proposed for the same purpose [68, 8], which mainly learn a good latent representation of the given class of data using various auto-encoders [1, 14, 65, 71, 83, 69] or GAN [27] style adversarial training [74, 72, 15, 64].”