Dear Colleagues,

We would like to invite you to the following SPASS seminar, jointly organized by UniPi, SNS, UniFi and UniSi (abstract below):

Scalable large scale variational inference

by Omiros Papaspiliopoulos (Università Bocconi)

The seminar will take place on TUE, 15.10.2024 at 14:00 CET in Aula Seminari, Dipartimento di Matematica, UNIPI and streamed online at this link.

The organizers,

A. Agazzi, G. Bet, A. Caraceni, F. Grotto, G. Zanco

https://sites.google.com/unipi.it/spass

--------------------------------------------

Title: Scalable large scale variational inference

Abstract: The motivation behind this work is a computational framework for approximate statistical inference for the broad class of generalized bi-linear mixed models, special cases of which are generalized linear mixed models, factor models and probabilistic factorization models to mention a few popular examples. The objective is that the resultant inference has theoretically provable guarantees in terms of both uncertainty quantification and computational complexity.

In this talk, I will first give some highlights of our work in using this framework for analyzing political ideology in Europe and then I will give a stand-alone short introduction to variational inference, discussing some theoretical aspects of its scalability.