0.25
0.5
0.75
1.25
1.5
1.75
2
Spoken Dialog Systems as an Application of Planning under Uncertainty
Published on Jul 21, 20115033 Views
Spoken dialog systems present a classic example of planning under uncertainty. In a spoken dialog system, a computer is trying to help a person accomplish something, using spoken language as the commu
Related categories
Chapter list
Spoken dialog systems as an application of planning under uncertainty00:00
What is a spoken dialog system?00:00
Speech recognition and spoken language understanding00:06
Speech recognition and spoken language understanding00:43
In-car spoken dialogue system01:34
Automated receptionist02:16
Outline03:25
Speech recognition and spoken language understanding04:10
Why speech recognition is hard: some examples04:52
ASR/SLU errors are common - 106:49
ASR/SLU errors are common - 207:16
ASR/SLU errors are common - 307:30
ASR errors are hard to detect07:50
ASR/SLU errors are common - 408:47
Curse of history - 109:06
Curse of history - 210:11
How dialog systems are built today - 110:43
How dialog systems are built today - 210:53
How dialog systems are built today - 311:30
Comercial System11:41
Problems12:08
Treating dialogue systems as planning under uncertainty12:56
Idea: track a distribution over dialogue states 13:08
Tracking a distribution over dialogue states - 114:47
Tracking a distribution over dialogue states - 215:02
Tracking a distribution over dialogue states - 315:06
Distribution update equation15:35
Problem: Updating belief in real-time16:07
2 methods for efficient belief monitoring16:36
M-Best partitions: Intuition - 116:41
M-Best partitions: Intuition - 217:04
Partition update equation17:17
Partition update example (maxPartitions = 3) - 117:56
Partition update example (maxPartitions = 3) - 218:30
Partition update example (maxPartitions = 3) - 318:38
Partition update example (maxPartitions = 3) - 418:42
Partition update example (maxPartitions = 3) - 518:47
Partition update example (maxPartitions = 3) - 618:49
Partition update example (maxPartitions = 3) - 719:06
Partition update example (maxPartitions = 3) - 819:13
Example - 119:32
Network-based approaches20:36
Example - 221:23
Tracking multiple dialogue states: example21:49
Tracking multiple dialogue states: results22:36
Treating dialogue systems as planning under uncertainty23:17
Idea: choose actions via a reward function23:22
A Simple Two State Example (voicemail)24:21
Policy Value Function at 30% Error Rate25:12
Policy Value Function vs Error Rate25:50
Planning : issues for real-world use26:31
Scaling up : what are the difficult decisions? - 126:45
Scaling up : what are the difficult decisions? - 227:11
Scaling up : what are the difficult decisions? - 327:28
Scaling up : what are the difficult decisions? - 427:35
Planning : Useful features for difficult decisions27:49
Domain knowledge & business rules - 128:11
Domain knowledge & business rules - 229:31
Example - 330:25
Example - 430:55
Reinforcement Learning: results32:04
Prospects for commercial use32:53
Tracking a distribution over dialog states32:58
Multiple dialog states33:46
Automatic action selection - 134:07
Automatic action selection - 234:43
Spoken Dialogue Challenge 201035:27
Example - 437:22
Some thoughts on the future39:13
Which planning algorithms to use?39:22
Feature selection in RL40:03
What is a good simulated user?40:44
What is a good reward function?41:44
In conclusion42:42
If you want to get started...43:32
Thanks to my collaborators44:10
Thanks!44:16