BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Gradient-based Adaptive Markov Chain Monte Carlo - Dr Michalis Tit
 sias\, DeepMind
DTSTART:20191203T150000Z
DTEND:20191203T160000Z
UID:TALK135397@talks.cam.ac.uk
CONTACT:Prof. Ramji Venkataramanan
DESCRIPTION:We introduce a gradient-based learning method to automatically
  adapt Markov chain Monte Carlo (MCMC) proposal distributions to intractab
 le targets. We define a maximum entropy regularised objective function\, r
 eferred to as generalised speed measure\, which can be robustly optimised 
 over the parameters of the proposal distribution by applying stochastic gr
 adient optimisation. An advantage of our method compared to traditional ad
 aptive MCMC methods is that the adaptation occurs even when candidate stat
 e values are rejected. This is a highly desirable property of any adaptati
 on strategy because the adaptation starts in early iterations even if the 
 initial proposal distribution is far from optimum. We apply the framework 
 for learning multivariate random walk Metropolis and Metropolis-adjusted L
 angevin proposals with full covariance matrices\, and provide empirical ev
 idence in high dimensional targets that our method can outperform other MC
 MC algorithms\, including Hamiltonian Monte Carlo schemes.\n\n\n*Bio*:\nMi
 chalis Titsias received a Diploma in Informatics from the University of Io
 annina\, Greece\, in 1999\, an MSc degree also from the University of Ioan
 nina\, in 2001\, and a PhD degree from the School of Informatics\, Univers
 ity of Edinburgh\, in 2005. From October 2007 to July 2011\, he worked as 
 a research associate in the machine learning and optimisation research gro
 up at the School of Computer Science of the University of Manchester\, whi
 le from November 2011 to September 2012 he worked as a postdoctoral resear
 ch scientist in statistical cancer genomics at the Wellcome Trust Centre f
 or Human Genetics and the Department of Statistics at the University of Ox
 ford. From 2012 to 2018 was firstly a Lecturer\, and later an Assistant Pr
 ofessor\,\nin the Department of Informatics of the Athens University of Ec
 onomics and Business\, Greece. From October 2018\, he works as a full time
  Research Scientist at DeepMind in London\, UK. His research interests inc
 lude machine learning\, deep learning\, reinforcement learning\, data scie
 nce and Bayesian statistics.
LOCATION:LR11\, Department of Engineering
END:VEVENT
END:VCALENDAR