# Bayesian Neural Network Priors Revisited Source: [Bayesian Neural Network Priors Revisited]() Collection index: [llms.txt]() Data updated: 2025-10-27T06:13:32.751Z Authors: Vincent Fortuin, Adri\\\`a Garriga-Alonso, Sebastian W. Ober, Florian Wenzel, Gunnar R\\"atsch, Richard E. Turner, Mark van der Wilk, Laurence Aitchison Published: 2021-02-12T15:18:06.000Z ## Abstract Isotropic Gaussian priors are the de facto standard for modern Bayesian neural network inference. However, it is unclear whether these priors accurately reflect our true beliefs about the weight distributions or give optimal performance. To find better priors, we study summary statistics of neural network weights in networks trained using stochastic gradient descent (SGD). We find that convolutional neural network (CNN) and ResNet weights display strong spatial correlations, while fully connected networks (FCNNs) display heavy-tailed weight distributions. We show that building these observations into priors can lead to improved performance on a variety of image classification datasets. Surprisingly, these priors mitigate the cold posterior effect in FCNNs, but slightly increase the cold posterior effect in ResNets. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Image Classification on Shift-MNIST (test)](): Image Classification; Shift-MNIST (test); Accuracy: 90.83. ## Original sources and implementation [Original paper]() [PDF]()