AI-Fun with ELLIS Seminar | José Miguel Hernández-Lobato: Deep Generative Models for Molecular Simulation
| Dates: | 2 October 2024 |
|---|
| Times: | 11:00 - 12:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Faculty of Science and Engineering |
|---|
| Who is it for: | University staff, External researchers, Alumni, Current University students |
|---|
| Speaker: | José Miguel Hernández-Lobato |
|---|
|
The Manchester Centre for AI Fundamentals and Manchester's ELLIS Unit are co-hosting a series of seminars featuring expert researchers working in the fundamentals of AI.
Abstract:
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 ?-divergence with ?=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.
José Miguel Hernández-Lobato short bio:
Since Sep 2016, I am a University Lecturer (equivalent to US Assistant Professor) in Machine Learning at the Department of Engineering in the University of Cambridge, UK. I was before a postdoctoral fellow in the Harvard Intelligent Probabilistic Systems group at the School of Engineering and Applied Sciencies of Harvard University, working with the group leader Prof. Ryan Adams. This position was funded through a post-doctoral fellowship given by the Rafael del Pino Foundation. Before that, I was a postdoctoral research associate in the Machine Learning Group at the Department of Engineering in the University of Cambridge (UK) from June 2011 to August 2014, working with Prof. Zoubin Ghahramani. During my first two years in Cambridge I worked in a collaboration project with the Indian multinational company Infosys Technologies. I also spent two weeks giving lectures on Bayesian Machine Learning at Charles University in Prague (Czech Republic). From December 2010 to May 2011, I was a teaching assistant at the Computer Science Department in Universidad Autónoma de Madrid (Spain), where I completed my Ph.D. and M.Phil. in Computer Science in December 2010 and June 2007, respectively. I also obtained a B.Sc. in Computer Science from this institution in June 2004, with a special prize to the best academic record on graduation. My research revolves around model based machine learning with a focus on probabilistic learning techniques and with a particular interest on Bayesian optimization, matrix factorization methods, copulas, Gaussian processes and sparse linear models. A general feature of my work is also an emphasis on fast methods for approximate Bayesian inference that scale to large datasets. The results of my research have been published at top machine learning journals (Journal of Machine Learning Research) and conferences (NIPS and ICML).
Speaker
José Miguel Hernández-Lobato
Travel and Contact Information
Find event
Emmeline Suite
Christabel Pankhurst Building
