The Deep Weight Prior

Abstract

Bayesian inference is known to provide a general framework for incorporating prior knowledge or specific properties into machine learning models via carefully choosing a prior distribution. In this work, we propose a new type of prior distributions for convolutional neural networks, deep weight prior (DWP), that exploit generative models to encourage a specific structure of trained convolutional filters e.g., spatial correlations of weights. We define DWP in the form of an implicit distribution and propose a method for variational inference with such type of implicit priors. In experiments, we show that DWP improves the performance of Bayesian neural networks when training data are limited, and initialization of weights with samples from DWP accelerates training of conventional convolutional neural networks.

arXiv commentTL;DR: The deep weight prior learns a generative model for kernels of convolutional neural networks, that acts as a prior distribution while training on new datasets

Cited by in corpus (11)

• Mathematical Models of Overparameterized Neural Networks
• Bayesian Neural Network Priors Revisited
• All You Need is a Good Functional Prior for Bayesian Deep Learning
• Multi-Task Variational Information Bottleneck
• A Bit More Bayesian: Domain-Invariant Learning with Uncertainty
• Importance Weighted Hierarchical Variational Inference
• Learning a smooth kernel regularizer for convolutional neural networks
• Sparse Uncertainty Representation in Deep Learning with Inducing Weights
• Doubly Semi-Implicit Variational Inference
• Symmetric Wasserstein Autoencoders
• Relational Weight Priors in Neural Networks for Abstract Pattern Learning and Language Modelling