Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples
Kimin Lee, Honglak Lee, Kibok Lee, Jinwoo Shin
TL;DR
This work tackles the problem of overconfident predictions in deep classifiers by training with a confidence loss that penalizes certainty on out-of-distribution samples and a boundary-focused GAN that generates informative OOD examples. The method jointly trains a classifier and a boundary-sampling GAN in alternating steps, aiming to maximize separation between in- and out-of-distribution outputs without sacrificing in-distribution accuracy. Empirically, the approach improves the effectiveness of threshold-based OOD detectors across standard image datasets, notably boosting true negative rates for LSUN on CIFAR-10 and SVHN. The proposed framework provides a practical guidance for deploying safer classifiers in real-world settings by calibration-aware training.
Abstract
The problem of detecting whether a test sample is from in-distribution (i.e., training distribution by a classifier) or out-of-distribution sufficiently different from it arises in many real-world machine learning applications. However, the state-of-art deep neural networks are known to be highly overconfident in their predictions, i.e., do not distinguish in- and out-of-distributions. Recently, to handle this issue, several threshold-based detectors have been proposed given pre-trained neural classifiers. However, the performance of prior works highly depends on how to train the classifiers since they only focus on improving inference procedures. In this paper, we develop a novel training method for classifiers so that such inference algorithms can work better. In particular, we suggest two additional terms added to the original loss (e.g., cross entropy). The first one forces samples from out-of-distribution less confident by the classifier and the second one is for (implicitly) generating most effective training samples for the first one. In essence, our method jointly trains both classification and generative neural networks for out-of-distribution. We demonstrate its effectiveness using deep convolutional neural networks on various popular image datasets.