CNS/EE 148 - Graphical models and applications

 

Syllabus

Introduction to graphical models (Polito)

Basics on graphical models and statistics

·         Basics of graph teory. Families of probability distributions associated to direcetd and undirected graphs.
Markov properties and conditional independence.

·         Statistical concepts as building blocks for graphical models. MAP and ML estimation.
Density estimation, classification and regression.

Learning from data

·         Model structure and parameters estimation.

·         Complete observations and latent variables.

·         The EM algorithm.

·         Model selection.

Exact inference

·         The junction tree and related algorithms. Belief propagation and belief revision.

·         The generalized distributive law.

·         Hidden Markov Models and Kalmann Filtering with graphical models.

Approximate inference

·         Variational methods.

·         Monte Carlo Methods.

·         Loopy junction graphs and loopy belief propagation.

·         Performance of loopy belief propagation.

·         Kullback-Leibler divergence and entropy. Bethe approximation of free energy and belief propagation.

Applications to Vision (Perona)

·         Bayesian Networks applied to a speech and visual recognition system.

·         Low level vision: inferring scene for image. Application to super-resolution, shading/reflectance variations, motion estimation.

·         Application to human motion detection. Labelling problem.

Applications to Coding Theory (McEliece)

Belief Propagation and Spin Glasses