A Bayesian Approach to Invariant Deep Neural Networks

07/20/2021

∙

We propose a novel Bayesian neural network architecture that can learn invariances from data alone by inferring a posterior distribution over different weight-sharing schemes. We show that our model outperforms other non-invariant architectures, when trained on datasets that contain specific invariances. The same holds true when no data augmentation is performed.

READ FULL TEXT

A Bayesian Approach to Invariant Deep Neural Networks

Split personalities in Bayesian Neural Networks: the case for full marginalisation

Neural Network Architecture for Database Augmentation Using Shared Features

Practical Bayesian Learning of Neural Networks via Adaptive Subgradient Methods

Masked Bayesian Neural Networks : Computation and Optimality

Data augmentation in Bayesian neural networks and the cold posterior effect

Correlated Parameters to Accurately Measure Uncertainty in Deep Neural Networks

Scene Uncertainty and the Wellington Posterior of Deterministic Image Classifiers

A Bayesian Approach to Invariant Deep Neural Networks

Related Research

Split personalities in Bayesian Neural Networks: the case for full marginalisation

Neural Network Architecture for Database Augmentation Using Shared Features

Practical Bayesian Learning of Neural Networks via Adaptive Subgradient Methods

Masked Bayesian Neural Networks : Computation and Optimality

Data augmentation in Bayesian neural networks and the cold posterior effect

Correlated Parameters to Accurately Measure Uncertainty in Deep Neural Networks

Scene Uncertainty and the Wellington Posterior of Deterministic Image Classifiers