An introduction to Bayesian inverse problems

schedule le lundi 16 décembre 2019 de 17h00 à 18h00

Organisé par : F. Bechtold, W. Da Silva , A. Fermanian, S. Has, Y. Yu

Intervenant : Kweku Abraham (LPSM)
Lieu : P6 Jussieu 16-26-209

Sujet : P6 Jussieu 16-26-209

Résumé :

In a statistical inverse problem you don't directly observe the quantity of interest, but its image under some forward map, further corrupted by noise. Recovering the quantity of interest should then involve inverting the forward map, which can be very difficult (think of the Calderon problem discussed a couple of weeks ago). Bayesian methods suggest a way round this, but do they work from a (frequentist) theoretical perspective? I'll explain why to expect that the answer is yes, and outline some conditions under which we can prove it. This shows that inverse problems are one setting where you don't have to philosophically prefer Bayesian ideas to benefit from using the methods.