This repository includes my own PyTorch implementations of a few variational inference algorithms for Bayesian neural networks:
- Bayes-by-Backprop (BBB) https://arxiv.org/pdf/1505.05424.pdf
- Matrix Variate Gaussian (MVG) https://arxiv.org/pdf/1603.04733.pdf
- Multiplicative Normalizing Flows (MNF) https://arxiv.org/pdf/1703.01961.pdf
- Bayes-by-Hypernet (BBH) https://arxiv.org/pdf/1711.01297.pdf
Unfortunately, the BBH implementation is not currently working. Example.ipynb shows each algorithm applied to a toy regression problem.