Learning Markov Random Fields and Models of Statistical Physics from Data

582753
5
Algorithms and machine learning
Advanced studies
The aim of this course is to explore the connection between Machine Learning and Statistical Physics.

Exam

18.12.2015 16.00 B123
Year Semester Date Period Language In charge
2015 autumn 28.10-31.12. 2-2 English Onur Dikmen

Lectures

Time Room Lecturer Date
Wed 14-16 B222 Onur Dikmen 28.10.2015-11.12.2015
Fri 14-16 B222 Onur Dikmen 28.10.2015-11.12.2015

Information for international students

The course will be taught in English.
 

General

Quick link: course Moodle page  (the access key will be communicated in the first lecture)
 
The course introduces undirected graphical models which are widely used in physics, biology, image processing, and many more disciplines, and teaches you how to estimate them from data.

 

Lecturers

 

Content

- Background on statistical inference (likelihood function, Bayes theorem ...)
- Ising models / Boltzmann machines and other Markov random fields
- Reasons why exact inference is difficult for Markov random fields 
- Approximate inference methods (Monte Carlo methods, pseudo-likelihood, variational Bayes, ...)
- Applications
- Deep learning
 

Prerequisites

A good understanding of Linear Algebra, Calculus and Probability. Familiarity with a programming language (e.g. Matlab, Python, R,...)
 

Grading

Regular exams: Homework 40%, exam 60%.
Separate exams: Homework 0%, exam 100%.
 

Material

Lecture notes are provided on Moodle.