TorchBayesian

Uncertainty-aware models in PyTorch.

Get Started Documentation

A practical library for uncertainty quantification and Bayesian neural networks in PyTorch, designed to integrate seamlessly into existing workflows.

• Easy to use
• Full compatibility with the PyTorch ecosystem
• Modular design that preserves PyTorch's flexibility
• Supports a wide range of methods
• Converts existing models easily

Tutorials

Regression

Predictive distributions and uncertainty intervals for continuous targets.

Classification

Uncertainty-aware prediction for classification tasks.

Segmentation

Dense prediction workflows with uncertainty-aware inference.

Methods

Bayes by Backprop

Variational Bayesian inference through learned posterior distributions.

Monte Carlo Dropout

Stochastic inference for predictive uncertainty estimation.

Priors and Posteriors

Flexible Bayesian parameterizations for uncertainty-aware models.