WAIC, but Why? Generative Ensembles for Robust Anomaly Detection

By Hyun-Jae Choi, Eric Jang, Alexander A. Alemi. 2018

289 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

VFlow: More Expressive Generative Flows with Variational Data Augmentation (2020, background): “, 2019), anomaly detection (Choi et al., 2018), and ray tracing (Müller et al.”
Energy-based Out-of-distribution Detection (2020, background): “These approaches are stronger baselines than pure generative-modeling-based OOD detection methods [4, 28, 33], due to the use of labeling information during training.”
Likelihood Ratios for Out-of-Distribution Detection (2019, methodology): “Our observations show a failure mode of deep generative models for OOD detection on genomic sequences and are complementary to earlier work which showed similar results for deep generative models on images (Nalisnick et al., 2018; Choi et al., 2018).”
CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (2020, background): “Several work thus proposed modiﬁed scores, e . g ., WAIC [7], likelihood ratio [52], and input complexity [57], or utilized unnormalized likelihood ( i . e ., energy) [12, 17].”
Generalized Out-of-Distribution Detection: A Survey (2021, unknown): “Recent methods turn to new scores such as likelihood regret [172] or an ensemble of multiple density models [168].”
Input complexity and out-of-distribution detection with likelihood-based generative models (2019, background): “While Choi et al. (2018) do not explicitly address the problem of typicality, their estimate of the Watanabe-Akaike information criterion using ensembles of generative models performs well in practice.”
Contrastive Training for Improved Out-of-Distribution Detection (2020, methodology): “[5] and likelihood ratios were estimated as an OOD scoring metric in [30].”
Efficient Learning of Generative Models via Finite-Difference Score Matching (2020, methodology): “As an additional evaluation of the learned model’s performance, we consider two tasks using deep EBMs: the first one is out-of-distribution detection, where we follow previous work [6, 43] to use typicality as the detection metric (details in Appendix C.”
Further Analysis of Outlier Detection with Deep Generative Models (2020, unknown): “However, previous efforts to utilize similar ideas for outlier detection have not been consistently successful [3, 11].”
Semantically Coherent Out-of-Distribution Detection (2021, methodology): “Existing OOD detection methods mostly focus on calibrating the distribution of the softmax layer [4] through temperature scaling [7], generative models [8, 9], or ensemble methods [10, 11].”
Detecting Out-of-Distribution Inputs to Deep Generative Models Using a Test for Typicality (2019, methodology): “For instance, Glow [22], a state-of-the-art normalizing ﬂow, trained on CIFAR-10 will assign a higher likelihood to SVHN than to its CIFAR-10 training data [26, 8].”
Anomalous Instance Detection in Deep Learning: A Survey (2020, background): “The key idea in [50] is that the likelihood models assign higher density values to the OOD examples than the in-distribution examples.”
Why Normalizing Flows Fail to Detect Out-of-Distribution Data (2020, methodology): “Point anomaly detection Choi et al. [7] proposed to estimate the Watanabe-Akaike Information Criterion using an ensemble of generative models, showing accurate OOD detection on some of the challenging dataset pairs.”
CutPaste: Self-Supervised Learning for Anomaly Detection and Localization (2021, background): “Anomaly detection under one-class classification setting, where we assume only the normal data is given during the training, has been widely studied [49, 56, 46, 63, 66, 13, 42, 36, 27].”