BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:RISE: Randomized Input Sampling for Explanation of Black-box Model
 s - Vitali Petsiuk\, Boston University
DTSTART:20180830T131500Z
DTEND:20180830T140000Z
UID:TALK109351@talks.cam.ac.uk
CONTACT:71965
DESCRIPTION:Deep neural networks are being used increasingly to automate d
 ata analysis and decision making\, yet their decision-making process is la
 rgely unclear and is difficult to explain to the end users. In this paper\
 , we address the problem of Explainable AI for deep neural networks that t
 ake images as input and output a class probability. We propose an approach
  called RISE that generates an importance map indicating how salient each 
 pixel is for the model's prediction. In contrast to white-box approaches t
 hat estimate pixel importance using gradients or other internal network st
 ate\, RISE works on black-box models. It estimates importance empirically 
 by probing the model with randomly masked versions of the input image and 
 obtaining the corresponding outputs. We compare our approach to state-of-t
 he-art importance extraction methods using both an automatic deletion/inse
 rtion metric and a pointing metric based on human-annotated object segment
 s. Extensive experiments on several benchmark datasets show that our appro
 ach matches or exceeds the performance of other methods\, including white-
 box approaches.
LOCATION:FW11 Meeting Room\, Computer Laboratory
END:VEVENT
END:VCALENDAR