BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Towards true end-to-end learning &amp\; optimization - Dr Frank Hu
 tter
DTSTART:20171117T110000Z
DTEND:20171117T120000Z
UID:TALK95422@talks.cam.ac.uk
CONTACT:Pat Wilson
DESCRIPTION:Deep neural networks automatically learn representations from 
 raw data\, but their architectures and hyperparameters still typically nee
 d to be defined manually by human experts. In this talk\, I will discuss e
 xtensions of Bayesian optimization for effectively searching in this combi
 ned space of architectures and hyperparameters\, thereby paving the way to
  fully automated machine learning (AutoML) with neural networks. I will fi
 rst show competition-winning practical AutoML systems and then focus on sp
 eeding up AutoML (sometimes up to 100-fold) by reasoning over data subsets
  and partial learning curves. I will also briefly show related application
 s to the end-to-end optimization of algorithms for solving hard combinator
 ial problems and discuss recent progress on weight optimization by variant
 s of stochastic gradient descent.
LOCATION:CBL Seminar Room
END:VEVENT
END:VCALENDAR