Deriving expectation maximization and variational inference algorithms from scratch.
Also check out my article: https://medium.com
"Variational Bayesian Inference for Unsupervised Clustering"
- Solved and released the first publicly available full mathematical derivation of a binomial mixture model given beta and Dirichlet priors using variational inference, including the exact evidence lower bound (ELBO)
- Implemented in PyTorch to speed matrix multiplication in fast, closed-form parameter update equations versus automatic differentiation and MCMC methods common in popular packages like PyMC and BayesPy