BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:What the DNN heard? Dissecting the machine brain for a better insi
 ght. -  Khe Chai Sim\, National University of Singapore
DTSTART:20151104T120000Z
DTEND:20151104T130000Z
UID:TALK62167@talks.cam.ac.uk
CONTACT:Anton Ragni
DESCRIPTION:Deep Neural Network (DNN) has been found to yield superior per
 formance compared to the conventional Gaussian Mixture Model (GMM) based s
 ystems for automatic speech recognition. However\, DNN has been used prett
 y much as a black box without much insights as to what the DNN has learned
 . This talk will present a novel approach for interpreting the DNN model\,
  which is based on analysing the hidden activity pattern. This technique c
 onstructs a 2-dimensional hidden activity space where interpretable region
 s can be defined. This technique can\nbe used to facilitate the understand
 ing and comparison of the hidden activity patterns across different hidden
  layers\, networks and time frames. Finally\, a technique called “stimul
 ated deep learning” will be presented\, where phone stimuli are used to 
 control the training process so that the hidden units of the resulting DNN
  yield interpretable activity patterns. Preliminary experimental results o
 n TIMIT show that the proposed stimulated deep learning technique is able 
 to learn DNNs with the hidden units showing phone-dependent activity regio
 ns without compromising the phone recognition performance.
LOCATION:Department of Engineering - LR5
END:VEVENT
END:VCALENDAR