A Bayesian Approach to Invariant Deep Neural Networks
arXiv:2107.09301
Abstract
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.
arXiv comment8 pages, 3 figures, To be published in ICML UDL 2021
References in corpus (14)
- Categorical Reparameterization with Gumbel-Softmax
- Weight Uncertainty in Neural Networks
- Probabilistic Backpropagation for Scalable Learning of Bayesian Neural Networks
- On the Generalization of Equivariance and Convolution in Neural Networks to the Action of Compact Groups
- A Comprehensive guide to Bayesian Convolutional Neural Network with Variational Inference
- Convolutional Gaussian Processes
- MDP Homomorphic Networks: Group Symmetries in Reinforcement Learning
- Learning Invariances using the Marginal Likelihood
- How Good is the Bayes Posterior in Deep Neural Networks Really?
- Bayesian Neural Network Priors Revisited
- BNNpriors: A library for Bayesian neural network inference with different prior distributions
- On the Benefits of Invariance in Neural Networks
- Repulsive Deep Ensembles are Bayesian
- On Stein Variational Neural Network Ensembles