BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium : Bayesian Machine Learning for Efficient Optimizati
 on of Black-box Functions
DTSTART:20160307T101500
DTEND:20160307T113000
DTSTAMP:20260407T063851Z
UID:32500f2b7341969acd6f0fb087787a7489036d2dbb6aa60890b724d8
CATEGORIES:Conferences - Seminars
DESCRIPTION:By : Jose Miguel Hernandez Lobato - Harvard School of Applied 
 Sciences and Engineering\nIC Faculty candidateAbstract :\nMany optimizatio
 n problems in engineering require trading off multiple objective functions
  that can only be evaluated through expensive calls to a black-box. Bayesi
 an optimization (BO) methods can solve these problems efficiently by perfo
 rming less function evaluations than other alternative approaches. To achi
 eve this\, BO methods combine 1) efficient data collection strategies with
  2) flexible probabilistic models and 3) accurate methods for approximate 
 inference. In this talk I will describe my most recent work in some of the
 se areas. First\, I will present new data collection strategies based on i
 nformation theory. Unlike previous methods\, these strategies can make opt
 imal decisions regarding the independent evaluation of different black-box
  objectives at different locations. I will illustrate the effectiveness of
  these methods in several optimization problems\, including the design of 
 optimal hardware accelerators for neural networks. In the second part of t
 he talk I will focus on using BO to speed up the discovery of optimal mole
 cules. To efficiently solve this problem we have to perform approximate in
 ference in very\ncomplicated models such as deep neural networks. For this
  purpose\, I will present "Black-box alpha"\, a new method for determinist
 ic inference that can be applied to such complex models with very little e
 ffort. Black-box alpha generalizes previous methods for deterministic infe
 rence (variational Bayes and expectation propagation) and can interpolate 
 between them by tuning a single parameter. This allows us to achieve excel
 lent results when making probabilistic predictions on molecule data with d
 eep neural networks.Bio :\nJose Miguel Hernandez Lobato is a postdoctoral 
 fellow in the Harvard School of Applied Sciences and Engineering since Sep
 tember 2014. His research interests\nare in Bayesian optimization\, scalab
 le methods for approximate inference and flexible probabilistic modeling o
 f data. Jose Miguel's research is driven by applications of machine learni
 ng to expensive optimal design problems in engineering. His work at Harvar
 d is currently funded by a grant from the Rafael del Pino Foundation. Befo
 re joining Harvard\, Jose Miguel was a postdoctoral research associate at 
 the Department of Engineering of Cambridge University (UK) were he worked 
 in a collaboration project with the Indian multinational company Infosys T
 echnologies. During this time Jose Miguel also gave lectures on Bayesian M
 achine Learning at Charles University in Prague (Czech Republic). From Dec
 ember 2010 to May 2011\, Jose Miguel was a teaching assistant at the Compu
 ter Science Department in Universidad Autonoma de Madrid (Spain)\, where h
 e obtained his Ph.D. and M. Phil. in Computer Science in December 2010 and
  June\n2007\, respectively. Jose Miguel also obtained a B.Sc. in Computer 
 Science from this institution in June 2004\, with a special prize to the b
 est academic record\non graduation.More information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR