Resumen/Abstract
Normalizing flows are tractable density models that can approximate complicated target distributions, e.g. Boltzmann distributions of physical systems. However, current methods for training flows either suffer from mode-seeking behavior, use samples from the target generated beforehand by expensive MCMC methods, or use stochastic losses that have high variance. To avoid these problems, we augment flows with annealed importance sampling (AIS) and minimize the mass-covering a-divergence with a=2, which minimizes importance weight variance. Our method, Flow AIS Bootstrap (FAB), uses AIS to generate samples in regions where the flow is a poor approximation of the target, facilitating the discovery of new modes. We apply FAB to multimodal targets and show that we can approximate them very accurately where previous methods fail. To the best of our knowledge, we are the first to learn the Boltzmann distribution of the alanine dipeptide molecule using only the unnormalized target density, without access to samples generated via Molecular Dynamics (MD) simulations: FAB produces better results than training via maximum likelihood on MD samples while using 100 times fewer target evaluations. After reweighting the samples, we obtain unbiased histograms of dihedral angles that are almost identical to the ground truth. Profesor Proponente EPS: Daniel Hernández Lobato
Curriculum ponente
Jose Miguel Hernandez Lobato is a lecturer in Machine Learning at the Department of Engineering of the University of Cambridge. Before that, he was a postdoctoral fellow at the Harvard School of Applied Sciences and Engineering from Sept. 2014 to Sept. 2016. His research interests are in Bayesian optimization, scalable methods for approximate inference, and flexible probabilistic modeling of data. Jose Miguel's research is driven by machine learning applications to expensive optimal design problems in engineering. Before joining Harvard, Jose Miguel was a postdoctoral research associate at the Department of Engineering of the University of Cambridge where he worked in a collaboration project with the Indian multinational company Infosys Technologies. From December 2010 to May 2011, Jose Miguel was a teaching assistant at the Computer Science Department at Universidad Autónoma de Madrid (Spain), where he obtained his Ph.D. and M.Phil. in Computer Science in December 2010 and June 2007, respectively. Jose Miguel also received a B.Sc. in Computer Science from this institution in June 2004, with a special prize for the best academic record on graduation.
Emisión por Teams: Seminario J.M Hernández Lobato 29/07/2024 | General | Microsoft Teams
Escuela Politécnica Superior | Universidad Autónoma de Madrid | Francisco Tomás y Valiente, 11 | 28049 Madrid | Tel.: +34 91 497 2222 | e-mail: informacion.eps@uam.es
