Tutorial: Bayes by Backprop¶

Notebook: BayesByBackprop_Tutorial.ipynb

Purpose¶

Train a Bayesian neural network with variational inference and analyze predictive uncertainty behavior in interpolation and extrapolation settings.

Data Setup¶

• Synthetic 1D nonlinear regression
• In-distribution and OOD evaluation ranges
• Controlled noise level to separate epistemic and aleatoric effects

Core Logic¶

• Bayesian linear layers with sampled weights
• ELBO optimization using vi_elbo_step
• MC-based predictive mean and confidence intervals

Expected Outputs¶

• ELBO/NLL/KL training curves
• fit quality on train/test
• uncertainty widening in OOD regions