Analysis of Confident-Classifiers for Out-of-distribution Detection

Abstract

Discriminatively trained neural classifiers can be trusted, only when the input data comes from the training distribution (in-distribution). Therefore, detecting out-of-distribution (OOD) samples is very important to avoid classification errors. In the context of OOD detection for image classification, one of the recent approaches proposes training a classifier called "confident-classifier" by minimizing the standard cross-entropy loss on in-distribution samples and minimizing the KL divergence between the predictive distribution of OOD samples in the low-density regions of in-distribution and the uniform distribution (maximizing the entropy of the outputs). Thus, the samples could be detected as OOD if they have low confidence or high entropy. In this paper, we analyze this setting both theoretically and experimentally. We conclude that the resulting confident-classifier still yields arbitrarily high confidence for OOD samples far away from the in-distribution. We instead suggest training a classifier by adding an explicit "reject" class for OOD samples.

arXiv commentSafeML 2019 ICLR workshop paper

References in corpus (7)

• A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
• Deep Anomaly Detection with Outlier Exposure
• Improving Reconstruction Autoencoder Out-of-distribution Detection with Mahalanobis Distance
• Denoising Autoencoders for Overgeneralization in Neural Networks
• Safer Classification by Synthesis
• Building robust classifiers through generation of confident out of distribution examples
• A randomized gradient-free attack on ReLU networks

Cited by in corpus (5)

• Out-of-distribution Detection in Classifiers via Generation
• SoK: Machine Learning Governance
• Anomalous Example Detection in Deep Learning: A Survey
• BaCOUn: Bayesian Classifers with Out-of-Distribution Uncertainty
• FROB: Few-shot ROBust Model for Classification and Out-of-Distribution Detection