Reverse KL-Divergence Training of Prior Networks: Improved Uncertainty and Adversarial Robustness

Abstract

Ensemble approaches for uncertainty estimation have recently been applied to the tasks of misclassification detection, out-of-distribution input detection and adversarial attack detection. Prior Networks have been proposed as an approach to efficiently \emph{emulate} an ensemble of models for classification by parameterising a Dirichlet prior distribution over output distributions. These models have been shown to outperform alternative ensemble approaches, such as Monte-Carlo Dropout, on the task of out-of-distribution input detection. However, scaling Prior Networks to complex datasets with many classes is difficult using the training criteria originally proposed. This paper makes two contributions. First, we show that the appropriate training criterion for Prior Networks is the \emph{reverse} KL-divergence between Dirichlet distributions. This addresses issues in the nature of the training data target distributions, enabling prior networks to be successfully trained on classification tasks with arbitrarily many classes, as well as improving out-of-distribution detection performance. Second, taking advantage of this new training criterion, this paper investigates using Prior Networks to detect adversarial attacks and proposes a generalized form of adversarial training. It is shown that the construction of successful \emph{adaptive} whitebox attacks, which affect the prediction and evade detection, against Prior Networks trained on CIFAR-10 and CIFAR-100 using the proposed approach requires a greater amount of computational effort than against networks defended using standard adversarial training or MC-dropout.

References in corpus (7)

• Efficient Estimation of Word Representations in Vector Space
• Auto-Encoding Variational Bayes
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Delving into Transferable Adversarial Examples and Black-box Attacks
• On Evaluating Adversarial Robustness
• Evidential Deep Learning to Quantify Classification Uncertainty
• On the (Statistical) Detection of Adversarial Examples

Cited by in corpus (11)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• Uncertainty in Gradient Boosting via Ensembles
• Uncertainty Estimation in Autoregressive Structured Prediction
• Information Aware Max-Norm Dirichlet Networks for Predictive Uncertainty Estimation
• Prior and Posterior Networks: A Survey on Evidential Deep Learning Methods For Uncertainty Estimation
• Regression Prior Networks
• DS-UI: Dual-Supervised Mixture of Gaussian Mixture Models for Uncertainty Inference
• Scaling Ensemble Distribution Distillation to Many Classes with Proxy Targets
• Shifts: A Dataset of Real Distributional Shift Across Multiple Large-Scale Tasks
• Real-time Policy Distillation in Deep Reinforcement Learning
• On Out-of-distribution Detection with Energy-based Models