README

Class Project: Alternating Direction Augmented Lagrangian for Probabilistic Semidefinite Models
Date: Dec 2010
Based on:
1. Crammer and Globerson. Discriminative learning via probabilistic semidefinite models. In: UAI, 2006.
2. Wen, Goldfarb, and Yin. Alternating Direction Augmented Lagrangian Methods for Semidefinite Programming. Mathematical Programming Computation, 2009.

The code implements the alternating direction methods in [2] to solve a 2-class version of the model in [1].
Code is online mainly for my reference. It is licensed under GPLv2 for non-commercial research purposes only.
The experiments were run on the Yale and AT&T faces datasets.