Adversarial Distillation of Bayesian Neural Network Posteriors
This reduces storage requirements for BNNs, enabling broader use in uncertainty-aware applications, though it is incremental as it builds on existing SGLD and GAN methods.
The paper tackles the high storage cost of Stochastic Gradient Langevin Dynamics (SGLD) for Bayesian neural networks by proposing Adversarial Posterior Distillation, which uses a GAN to distill SGLD samples, achieving no loss in performance on applications like anomaly detection, active learning, and defense against adversarial attacks.
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.