BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Constrained Bayesian optimization for automatic chemical design us
 ing variational autoencoders - Ryan-Rhys Griffiths
DTSTART:20200217T163000Z
DTEND:20200217T170000Z
UID:TALK139594@talks.cam.ac.uk
CONTACT:Bingqing Cheng
DESCRIPTION:Automatic Chemical Design is a framework for generating novel 
 molecules with optimized properties. The original scheme\, featuring Bayes
 ian optimization over the latent space of a variational autoencoder\, suff
 ers from the pathology that it tends to produce invalid molecular structur
 es. First\, we demonstrate empirically that this pathology arises when the
  Bayesian optimization scheme queries latent space points far away from th
 e data on which the variational autoencoder has been trained. Secondly\, b
 y reformulating the search procedure as a constrained Bayesian optimizatio
 n problem\, we show that the effects of this pathology can be mitigated\, 
 yielding marked improvements in the validity of the generated molecules. W
 e posit that constrained Bayesian optimization is a good approach for solv
 ing this kind of training set mismatch in many generative tasks involving 
 Bayesian optimization over the latent space of a variational autoencoder.
LOCATION:Mott Seminar (531) room\, top floor of the Mott Building\, in the
  Cavendish Laboratory\, West Cambridge.
END:VEVENT
END:VCALENDAR