We are happy to invite you to join the course “Graphical Models of Statistical Inference” by Prof. Michael Chertkov (Skoltech Adjunct Professor) planned for Term 1A.
When: 13/15/17 September, 5 PM
Where: IITP RUS (Kharkevich Institute) Bolshoy Karetny per. 19, build.1, Moscow
This course is recommended for IT students, as well as other specialization students, interested in learning about modern theoretical and practical approaches to analysis of big data sets with reach statistical correlations expressed through graphs, matrices, tensors and related. The course is light on rigorous proofs, but rich on statistics and physics intuition.
This mini-course will consist of the following six lectures:
- Graphical Models (Language) and Structured Statistical Inference (problem formulations) in Computer Science, Information Theory and Physics (intro).
- Computational Complexity & Algorithms (Deterministic & Stochastic). Statistical Inference as an Optimization — from Partition Function and Marginal Probabilities to Free Energy (Kublack-Leibler Functional).
- Mean-Field, Belief Propagation, Linear Programming — Variational Approaches, Relaxations, Lower and Upper Bounds. Exact & Heuristic approaches. Iterative Algorithms.
- Modern Analysis and Algorithmic Tools. Review of Loop Series, Cummulant Expansions, Computational Trees, Graph Cover & Monte-Carlo Approaches.
- Examples of Tractable Graphical Models: (a) Network Flows; (b) Attractive (Ferromagnetic) Ising Models; (c) Matching Models; (d) Planar (det-reducable) Models; (e) Gaussian Graphical Models.
- Open Problems. Various Applications, e.g. in Machine Learning, Energy, Bio and Social Systems. Connections/links to other areas of research in modern theoretical engineering.
This is an advanced level course suitable for second year M.Sc. and Ph.D. students. Some prior experience in Probability Theory, Statistics, Statistical Mechanics or Machine Learning (at least one credited course) is recommended.
For registration, please contact Skoltech Education Ofﬁce at