Bayesian Deep Learning

The Bayesian module provides methods for approximate Bayesian inference in neural networks, enabling uncertainty decomposition into epistemic and aleatoric components.

Methods

BaseBayesianMethod(*args, **kwargs)	Abstract base class for Bayesian deep learning methods.
MCDropout(model[, num_samples, dropout_rate])	Monte Carlo Dropout for uncertainty estimation.
DeepEnsemble(model_fn[, num_models])	Deep Ensembles for uncertainty quantification.
SWAG(model[, num_samples, var_clamp])	Stochastic Weight Averaging - Gaussian (SWAG).
LaplaceApproximation(model[, likelihood, ...])	Laplace Approximation for Bayesian Neural Networks.
VariationalBayesNN(in_features, ...[, ...])	Variational Bayesian Neural Network (Bayes by Backprop).
GaussianLinear(in_features, out_features[, ...])	Linear layer with Gaussian weights for variational inference.

Metrics

ensemble_diversity(predictions[, metric])	Measure diversity among ensemble members.
uncertainty_quality(uncertainties, errors)	Evaluate quality of uncertainty estimates.
disagreement(predictions[, method])	Compute disagreement score for each sample.
negative_log_likelihood(predictions, labels)	Compute negative log-likelihood of predictions.
brier_score(predictions, labels)	Compute Brier score for probabilistic predictions.
predictive_log_likelihood(predictions, labels)	Compute predictive log-likelihood (averaged over ensemble).
sharpness(predictions)	Compute sharpness of probabilistic predictions.

Utilities

predictive_entropy(predictions)	Compute predictive entropy for batched Bayesian predictions (total uncertainty).
mutual_information(predictions)	Compute mutual information (epistemic uncertainty).
expected_calibration_error(predictions, labels)	Compute Expected Calibration Error for Bayesian predictions.
decompose_uncertainty(predictions)	Decompose predictive uncertainty into epistemic and aleatoric components.
compute_disagreement(predictions)	Compute disagreement among ensemble members.
sample_from_posterior(mean, variance[, ...])	Sample from a Gaussian posterior.
ensemble_predictions_to_distribution(predictions)	Convert ensemble predictions to mean and variance.