Implicit Weight Uncertainty in Neural Networks

TL;DR

This paper introduces Bayes by Hypernet (BbH), a scalable variational approximation method using hypernetworks as implicit distributions, improving uncertainty estimation and robustness in neural networks.

Contribution

The paper proposes BbH, a novel scalable variational inference approach employing hypernetworks as implicit distributions for better uncertainty modeling.

Findings

Achieves competitive accuracy on MNIST and CIFAR5.

Provides more meaningful uncertainty estimates.

Demonstrates increased robustness against adversarial attacks.

Abstract

Modern neural networks tend to be overconfident on unseen, noisy or incorrectly labelled data and do not produce meaningful uncertainty measures. Bayesian deep learning aims to address this shortcoming with variational approximations (such as Bayes by Backprop or Multiplicative Normalising Flows). However, current approaches have limitations regarding flexibility and scalability. We introduce Bayes by Hypernet (BbH), a new method of variational approximation that interprets hypernetworks as implicit distributions. It naturally uses neural networks to model arbitrarily complex distributions and scales to modern deep learning architectures. In our experiments, we demonstrate that our method achieves competitive accuracies and predictive uncertainties on MNIST and a CIFAR5 task, while being the most robust against adversarial attacks.