Deep Anomaly Detection with Outlier Exposure

By Dan Hendrycks, Mantas Mazeika, Thomas G. Dietterich. 2018

1,782 citations recorded by the source corpus.

It is important to detect anomalous inputs when deploying machine learning systems. The use of larger and more complex inputs in deep learning magnifies the difficulty of distinguishing between anomalous and in-distribution examples. At the same time, diverse image and text data are available in enormous quantities. We propose leveraging these data to improve deep anomaly detection by training anomaly detectors against an auxiliary dataset of outliers, an approach we call Outlier Exposure (OE). This enables anomaly detectors to generalize and detect unseen anomalies. In extensive experiments on natural language processing and small- and large-scale vision tasks, we find that Outlier Exposure significantly improves detection performance. We also observe that cutting-edge generative models trained on CIFAR-10 may assign higher likelihoods to SVHN images than to CIFAR-10 images; we use OE to mitigate this issue. We also analyze the flexibility and robustness of Outlier Exposure, and identify characteristics of the auxiliary dataset that improve performance.

Subfields: Artificial Intelligence, Computer Networks and Communications, Safety, Robustness & Interpretability, Signal Processing

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

In Computer Networks and Communications

In Safety, Robustness & Interpretability

In Signal Processing

Citations in the corpus

Pretrained Transformers Improve Out-of-Distribution Robustness (2020, methodology): “Without access to an outlier dataset (Hendrycks et al., 2019b), the state-of-”
Measuring Massive Multitask Language Understanding (2020, background): “Another calibration measure is the Root Mean Squared (RMS) calibration error (Hendrycks et al., 2019a; Kumar et al., 2019).”
Natural Adversarial Examples (2019, background): “For out-of-distribution (OOD) detection [27, 39, 28, 29] models learn a distribution, such as the ImageNet-1K distribution, and are tasked with producing quality anomaly scores that distinguish between usual test set examples and examples from held-out anomalous distributions.”
Benchmarking Neural Network Robustness to Common Corruptions and Perturbations (2019, background): “…has focused on the important challenges of robustness to adversarial examples (Szegedy et al., 2014; Carlini & Wagner, 2017; 2016), unknown unknowns (Hendrycks et al., 2019; Hendrycks & Gimpel, 2017b; Liu et al., 2018), and model or data poisoning (Steinhardt et al., 2017; Hendrycks et al., 2018).”
Unsolved Problems in ML Safety (2021, background): “One of the symptoms of this problem is that anomaly detectors for large-scale images still cannot reliably detect that previously unseen random noise is anomalous [81].”
Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift (2018, background): “[20] expose the model to OOD samples, exploring heuristics for discriminating between in-distribution and out-of-distribution samples.”
Using Pre-Training Can Improve Model Robustness and Uncertainty (2019, methodology): “In the problem of out-of-distribution detection (Hendrycks & Gimpel, 2017b; Hendrycks et al., 2019; Lee et al., 2018a;b; Liu et al., 2018), models are tasked with (Chen et al., 2018).”
Continual Learning via Local Module Composition (2021, background): “Several recent works have shown that deep generative models often mistakenly assign high likelihood values to outlier points [68, 36, 95].”
Being Bayesian, Even Just a Bit, Fixes Overconfidence in ReLU Networks (2020, methodology): “Inspired by Hendrycks et al. (2019) and Hein et al. (2019), we simultaneously prefer high conﬁdence on the in-distribution validation set and low conﬁdence (high entropy) on the out-of-distribution validation set.”
Towards neural networks that provably know when they don't know (2019, methodology): “OE: Outlier exposure (Hendrycks et al., 2019) enforces uniform conﬁdence on a large out-distribution.”
Verified Uncertainty Calibration (2019, methodology): “Recent work in machine learning typically estimates each term in the calibration error directly from samples [7, 15, 16, 17].”
Out-of-Domain Detection for Natural Language Understanding in Dialog Systems (2019, methodology): “many contexts with different alias, such as “anomaly detection” [15], “one-class classification” [18], [19], “open-set recognition” [20], or “novelty detection” [21].”
Why ReLU Networks Yield High-Confidence Predictions Far Away From the Training Data and How to Mitigate the Problem (2018, methodology): “samples from p out on which one enforces a uniform distribution over the labels. We call this process in the following conﬁdence enhancing data augmentation (CEDA). We note that in a concurrent paper [16] a similar scheme has been proposed, where they use as p out existing large image datasets, wher”
Do Deep Generative Models Know What They Don't Know? (2018, background): “Yet, deep generative models can detect out-of-distribution inputs when using alternative metrics (Choi & Jang, 2018) and modified training procedures (Hendrycks et al., 2019).”
Energy-based Out-of-distribution Detection (2020, methodology): “Moreover, while JEM only utilizes in-distribution data, our framework allows exploiting both the in-distribution and the auxiliary outlier data to shape the energy gap ﬂexibly between the training and OOD data, a learning method that is much more effective than JEM or Outlier Exposure [15].”