Deep Anomaly Detection with Outlier Exposure

Abstract

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.

ICLR 2019; PyTorch code available at https://github.com/hendrycks/outlier-exposure

Cited by in corpus (25)

• AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
• Deep Nearest Neighbor Anomaly Detection
• Contrastive Training for Improved Out-of-Distribution Detection
• Out-of-distribution Detection in Classifiers via Generation
• Deep CNN-based Multi-task Learning for Open-Set Recognition
• Detecting Out-of-Distribution Examples with In-distribution Examples and Gram Matrices
• On Last-Layer Algorithms for Classification: Decoupling Representation from Uncertainty Estimation
• Unsupervised Out-of-Distribution Detection with Batch Normalization
• Novelty Detection Via Blurring
• Real-time Out-of-distribution Detection in Learning-Enabled Cyber-Physical Systems
• Analysis of Confident-Classifiers for Out-of-distribution Detection
• A Critical Evaluation of Open-World Machine Learning
• Better the Devil you Know: An Analysis of Evasion Attacks using Out-of-Distribution Adversarial Examples
• Efficacy of Pixel-Level OOD Detection for Semantic Segmentation
• Soft Labeling Affects Out-of-Distribution Detection of Deep Neural Networks
• Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector
• Mobile Traffic Classification through Physical Channel Fingerprinting: a Deep Learning Approach
• A Simple Framework for Uncertainty in Contrastive Learning
• Self-supervised Robust Object Detectors from Partially Labelled Datasets
• Bigeminal Priors Variational auto-encoder
• Deep Anomaly Detection by Residual Adaptation
• Likelihood Ratios and Generative Classifiers for Unsupervised Out-of-Domain Detection In Task Oriented Dialog
• Likelihood Assignment for Out-of-Distribution Inputs in Deep Generative Models is Sensitive to Prior Distribution Choice
• Towards Interpretable and Learnable Risk Analysis for Entity Resolution
• Why Should we Combine Training and Post-Training Methods for Out-of-Distribution Detection?