variational-inference

Overview

This project works through coordinate-ascent variational inference for a univariate normal model with unknown mean and variance. The R code keeps each parameter update visible and records the evidence lower bound at every iteration.

A written derivation accompanies the implementation, so the equations can be matched directly to the update steps in the code.

Technical highlights

• Mean-field Normal and Inverse-Gamma approximation
• Closed-form coordinate updates
• ELBO convergence tracking
• Posterior-density visualization

Topics