BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Global model explainability via aggregation - Umang Bhatt\, CMU
DTSTART:20190404T150000Z
DTEND:20190404T154500Z
UID:TALK122491@talks.cam.ac.uk
CONTACT:Adrian Weller
DESCRIPTION:Current approaches for explaining machine learning models fall
  into two distinct classes: antecedent event influence and value attributi
 on. The former leverages training instances to describe how much influence
  a training point exerts on a test point\, while the latter attempts to at
 tribute value to the features most pertinent to a given prediction. In thi
 s talk\, I will discuss my work\, *AVA: Aggregate Valuation of Antecedents
 *\, that fuses these two explanation classes to form a new approach to fea
 ture attribution that not only retrieves local explanations but also captu
 res global patterns learned by a model. We find that aggregating and weigh
 ting Shapley value explanations via AVA results in a valid Shapley value e
 xplanation. I will provide a medical use case for AVA explanations\, mirro
 ring diagnostic approaches used by healthcare professionals.\n\nI will als
 o discuss new heuristics and show preliminary results for aggregating loca
 l explanations from different explanation techniques using a _wisdom of th
 e crowds_ approach subject to a user specified criterion.  
LOCATION:Engineering Department\, CBL Room BE-438.
END:VEVENT
END:VCALENDAR