Tutorial: SGLD¶

Notebook: SGLD_Tutorial.ipynb

Purpose¶

Approximate Bayesian posterior sampling through stochastic gradient Langevin dynamics.

Data Setup¶

• Supervised learning setup with mini-batches
• Burn-in and sampling windows

Core Logic¶

• run SGLDOptimizer updates
• collect model parameter snapshots
• estimate predictive mean/variance from sampled models

Expected Outputs¶

• posterior sample trajectories
• uncertainty-aware predictions from sample aggregation