The article presents BayesDLL, a new Bayesian neural network library for PyTorch. The library is designed for large-scale deep networks and implements various approximate Bayesian inference algorithms such as variational inference, MC-dropout, stochastic-gradient MCMC, and Laplace approximation. It is user-friendly requiring virtually no code modifications. The library also allows pre-trained model weights to serve as a prior mean, beneficial for performing Bayesian inference with large-scale foundation models. BayesDLL is publicly available on GitHub.
Publication date: 22 Sep 2023
Project Page: https://github.com/SamsungLabs/BayesDLL1
Paper: https://arxiv.org/pdf/2309.12928
