Adversarial Distillation of Bayesian Neural Network Posteriors

Abstract

Bayesian neural networks (BNNs) allow us to reason about uncertainty in a principled way. Stochastic Gradient Langevin Dynamics (SGLD) enables efficient BNN learning by drawing samples from the BNN posterior using mini-batches. However, SGLD and its extensions require storage of many copies of the model parameters, a potentially prohibitive cost, especially for large neural networks. We propose a framework, Adversarial Posterior Distillation, to distill the SGLD samples using a Generative Adversarial Network (GAN). At test-time, samples are generated by the GAN. We show that this distillation framework incurs no loss in performance on recent BNN applications including anomaly detection, active learning, and defense against adversarial attacks. By construction, our framework not only distills the Bayesian predictive distribution, but the posterior itself. This allows one to compute quantities such as the approximate model variance, which is useful in downstream tasks. To our knowledge, these are the first results applying MCMC-based BNNs to the aforementioned downstream applications.

arXiv commentaccepted at ICML 2018

References in corpus (14)

• Striving for Simplicity: The All Convolutional Net
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• On Calibration of Modern Neural Networks
• Improved Training of Wasserstein GANs
• Improved Techniques for Training GANs
• Weight Uncertainty in Neural Networks
• Deep Bayesian Active Learning with Image Data
• Bayesian Active Learning for Classification and Preference Learning
• Stochastic Gradient Hamiltonian Monte Carlo
• Variational Approaches for Auto-Encoding Generative Adversarial Networks
• Bayesian Dark Knowledge
• Improving GANs Using Optimal Transport
• Adversarial Phenomenon in the Eyes of Bayesian Deep Learning
• Approximate Inference with Amortised MCMC

Cited by in corpus (14)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• Hands-on Bayesian Neural Networks -- a Tutorial for Deep Learning Users
• Your Classifier is Secretly an Energy Based Model and You Should Treat it Like One
• PID-GAN: A GAN Framework based on a Physics-informed Discriminator for Uncertainty Quantification with Physics
• Laplace Redux -- Effortless Bayesian Deep Learning
• HyperGAN: A Generative Model for Diverse, Performant Neural Networks
• Inhibited Softmax for Uncertainty Estimation in Neural Networks
• Infinitely Deep Bayesian Neural Networks with Stochastic Differential Equations
• PriMask: Cascadable and Collusion-Resilient Data Masking for Mobile Cloud Inference
• Ensemble Model Patching: A Parameter-Efficient Variational Bayesian Neural Network
• Generalized Bayesian Posterior Expectation Distillation for Deep Neural Networks
• URSABench: Comprehensive Benchmarking of Approximate Bayesian Inference Methods for Deep Neural Networks
• Assessing the Robustness of Bayesian Dark Knowledge to Posterior Uncertainty
• Efficient and Transferable Adversarial Examples from Bayesian Neural Networks