On-line Learning of Wide-domain Spoken Dialogue Systems
Jul 31, 2016 1160 views 
en-es
en-fr
en-sl
en
0.25
0.5
0.75
1.25
1.5
1.75
2
The first part of the talk reviews the general structure of limited domain statistical SDS and then explains how a collection of limited domain systems can be merged using the framework of Bayesian Committee Machines. The problem of reward estimation in on-line learning is then introduced and a solution based on the joint estimation of Gaussian Process based reward prediction and dialogue policy is presented.