External Lecturer:
Jean Barbier
Course Type:
PhD Course
Academic Year:
2020-2021
Period:
November - December
Duration:
26 h
Description:
10 x 2 hours + 2 x 3 hrs Labs = 26 hours (5 weeks: 04/11 - 04/12)
- Bayesian inference, information theory and statistical mechanics:
i. Statistical inference, Bayes formula and decision theory
ii. Surprise, Shannon entropy and mutual information
iii. Statistical mechanics of disordered systems 101, and links with Bayesian inference
iv. Lab 1 - Information-theoretic limits
i. Replica symmetric formula for the mutual information
ii. A powerful (exact) heuristic: the replica method
iii. Why ensembles matter? Concentration of the free energy
iv. Replica symmetry in inference: overlap concentration
v. Rigorous approach 1: the (adaptive) interpolation method
vi. Rigorous approach 2: the cavity method
vii. Lab 2 - Algorithmic limits
i. Message-passing
ii. State evolution, and optimality of approximate message-passing
Please, notice that this is a course belonging to Data Science Excellence Department programme. MAMA PhD students can plan 33% of their credits (i.e. 50 hrs) from this programme.
Research Group:
Location:
A-128