Luc Rey-Bellet: Variational representation and concentration inequalities for uncertainty quantification
Tid: Må 2020-10-26 kl 15.15 - 16.15
Föreläsare: Luc Rey-Bellet, UMass Amherst
We use variational representation for information-theoretic divergences (such as the KL-divergences) and concentration
inequalities to derive, in a systematic way, tight information inequalities. A central question in uncertainty quantification and statistical learning is how to choose the right divergences for the right quantity of interest. We concentrate on KL-divergence, f-divergences, and Renyi divergences and discuss several examples: expected values, variance and other functionals, rare events, and steady state expectation for MCMC dynamics (for example Hamiltonian Monte-Carlo).
Zoom notes: The passcode for this meeting is 321777. This meeting ID — 621 4469 8204 — will be the recurring meeting for the Statistics and Probability Seminar.