BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Grammar Variational Autoencoder - Matt Kusner\, ATI and Warwick Un
 iversity
DTSTART:20180220T140000Z
DTEND:20180220T150000Z
UID:TALK95530@talks.cam.ac.uk
CONTACT:Damon Wischik
DESCRIPTION:Deep generative models have been wildly successful at learning
  coherent latent representations for continuous data such as natural image
 s\, artwork\, and audio. However\, generative modeling of discrete data su
 ch as arithmetic expressions and molecular structures still poses signific
 ant challenges. Crucially\, state-of-the-art methods often produce outputs
  that are not valid.\n\nWe make the key observation that frequently\, disc
 rete data can be represented as a parse tree from a context-free grammar. 
 We propose a variational autoencoder which directly encodes from and decod
 es to these parse trees\, ensuring the generated outputs are always syntac
 tically valid. Surprisingly\, we show that not only does our model more of
 ten generate valid outputs\, it also learns a more coherent latent space i
 n which nearby points decode to similar discrete outputs. We demonstrate t
 he effectiveness of our learned models by showing their improved performan
 ce in Bayesian optimization for symbolic regression and molecule generatio
 n.
LOCATION:Centre for Mathematical Sciences\, MR4
END:VEVENT
END:VCALENDAR