Authors: Roger Fan rogerfan@umich.edu, Joseph Dickens josephdi@umich.edu

Stochastic Expectation-Maximization (SEM), a novel optimization algorithm based on simulated annealing and the Expectation-Maximization (EM) algorithm, is described and implemented for Gaussian mixture models. Also includes code for comparing SEM to traditional EM and Deterministic Annealing EM, two of the most common existing methods for estimating mixture models.

This code is open-sourced under the Mozilla Public License, version 2.0. See LICENSE for details.

This was a final project for Stats 608A, Fall 2015 at the University of Michigan, taught by Professor Ambuj Tewari.