BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning to Act in Noisy Contexts using Deep Proxy Learning - Arth
 ur Gretton (University College London)
DTSTART:20250509T130000Z
DTEND:20250509T140000Z
UID:TALK230509@talks.cam.ac.uk
DESCRIPTION:We consider the problem of evaluating the expected outcome of 
 an action or policy\, using off-policy observations\, where the relevant c
 ontext is noisy/anonymized. This scenario might arise due to privacy const
 raints\, data bandwidth restrictions\, or both. As an example\, users migh
 t wish to determine the anticipated outcome of an exercise regime\, with o
 nly an incomplete view available of their fitness levels (for instance\, f
 rom journaling or wearables). We will employ the recently developed tool o
 f proxy causal learning to address this problem. In brief\, two noisy view
 s of the context are used: one prior to the user action\, and one subseque
 nt to it\, and influenced by the action. This pair of views will allow us 
 to provably recover the average causal effect of an action under reasonabl
 e assumptions. As a key benefit of the proxy approach\, we need never expl
 icitly model or recover the hidden context. Our implementation employs lea
 rned neural net representations for both the action and context\, allowing
  each to be complex and high dimensional (images\, text). We demonstrate t
 he deep proxy learning method in a setting where the action is an image\, 
 and show that we outperform an autoencoder-based alternative.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR