Cautious Deep Learning

Abstract

Most classifiers operate by selecting the maximum of an estimate of the conditional distribution $p(y|x)$ where $x$ stands for the features of the instance to be classified and $y$ denotes its label. This often results in a {\em hubristic bias}: overconfidence in the assignment of a definite label. Usually, the observations are concentrated on a small volume but the classifier provides definite predictions for the entire space. We propose constructing conformal prediction sets which contain a set of labels rather than a single label. These conformal prediction sets contain the true label with probability $1-\alpha$. Our construction is based on $p(x|y)$ rather than $p(y|x)$ which results in a classifier that is very cautious: it outputs the null set --- meaning "I don't know" --- when the object does not resemble the training examples. An important property of our approach is that adversarial attacks are likely to be predicted as the null set or would also include the true label. We demonstrate the performance on the ImageNet ILSVRC dataset and the CelebA and IMDB-Wiki facial datasets using high dimensional features obtained from state of the art convolutional neural networks.

References in corpus (2)

• A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
• Least Ambiguous Set-Valued Classifiers with Bounded Error Levels

Cited by in corpus (9)

• Testing for Outliers with Conformal p-values
• Deep Gamblers: Learning to Abstain with Portfolio Theory
• Uncertainty Sets for Image Classifiers using Conformal Prediction
• Conformal prediction with localization
• Knowing what you know: valid and validated confidence sets in multiclass and multilabel prediction
• Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Control
• Cryptocurrency Price Prediction and Trading Strategies Using Support Vector Machines
• Private Prediction Sets
• Combining Deep Learning and Verification for Precise Object Instance Detection