CVN+MICS: Victor Elvira - TBA

Abstract

Importance sampling is a widely used family of Monte Carlo methods. It provides a simple and flexible framework for approximating expectations using samples from a proposal distribution. In this talk, we will start with the basic principles of importance sampling and self-normalized importance sampling, emphasizing the central role of proposal design in estimator efficiency. We will then review settings in which several proposal distributions are available, leading to multiple importance sampling. We will also revisit proposal design for self-normalized importance sampling, discussing recent results. Finally, we will discuss adaptive importance sampling, where proposal distributions are iteratively improved using information from previous samples, including modern population-based adaptation schemes.

Bio

Victor Elvira is Distinguished Professor at the Basque Centre for Applied Mathematics (Bilbao, Spain) and Professor of Statistics and Data Science at the University of Edinburgh (UK), where he previously also served as Director of the Centre for Statistics. He received his PhD in Statistical Signal Processing from the University of Cantabria (Spain). His research focuses on computational statistics, statistical signal processing, and probabilistic machine learning, with particular emphasis on Bayesian inference and Monte Carlo methods. He is an ELLIS Fellow and IEEE Senior Member, and has held Fulbright, Marie Curie, Leverhulme, and Alan Turing fellowships. In 2024, he received the EURASIP Early Career Award for contributions to importance sampling and particle filtering.