baetorch

Python library for Bayesian Autoencoders

Features

• Quantify epistemic uncertainty using approximate Bayesian inference
  • MC-Dropout
  • Bayesian Ensembling (with Anchored priors)
  • Variational Inference (Bayes by Backprop)
• Options for specifying data likelihood p(X|theta) to Gaussian or Bernoulli
• Quantify (homo/heteroskedestic) aleatoric uncertainty using Gaussian Likelihood
• Automatic learning rate finder for Bayesian Autoencoders

Evaluation and plots

Samples from BAE-Ensemble (M=5), FashionMNIST vs MNIST

Comparison of performance

Automatic learning rate finder (based on Leslie Smith's methods)

TODO:

• separate examples for VI, MCDropout, Ensemble
• show how to:
  • change fully-dense to convolutional autoencoders
  • use with/without learning rate finder
  • swap homoscedestic mode
  • run test suite