A Unifying Perspective of Parametric Policy Search Methods for Markov Decision Processes thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

A Unifying Perspective of Parametric Policy Search Methods for Markov Decision Processes

Published on Jan 16, 20132823 Views

Parametric policy search algorithms are one of the methods of choice for the optimisation of Markov Decision Processes, with Expectation Maximisation and natural gradient ascent being considered the c

Related categories

Chapter list

A Unifying Perspective of Parametric Policy Search Methods for Markov Decision Processes00:00
Outline00:04
Introduction00:06
Markov Decision Processes - Examples00:07
Notation00:46
MDP Solution Methods (1)02:47
MDP Solution Methods (2)03:09
MDP Solution Methods (3)03:16
MDP Solution Methods (4)03:18
MDP Solution Methods (5)03:20
MDP Solution Methods (6)03:24
Parametric Policy Search Methods03:36
Steepest Gradient Ascent (1)04:08
Steepest Gradient Ascent (2)04:48
Expectation Maximisation05:02
Contributions (1)05:43
Contributions (2)06:31
Contributions (3)06:57
Contributions (4)07:03
Analysis of Parametric Policy Search Methods07:22
Analysis - Overview (1)07:24
Analysis - Overview (2)07:31
Natural Gradient Ascent - Analysis08:05
Expectation Maximisation - Analysis08:47
Analysis Summary09:34
Approximate Newton Methods10:02
Approximate Newton Methods (1)10:10
Approximate Newton Methods - Properties (1)10:37
niApproximate Newton Methods - Properties (2)11:21
Approximate Newton Methods - Properties (3)12:07
Experiments12:33
Tetris Experiment12:33
Tetris - Experiment 112:43
Tetris - Experiment 213:19
Linear System Experiment (1)13:46
Linear System Experiment (2)13:49
Non-Linear System Experiment (1)14:08
Non-Linear System Experiment (1)14:15
Summary & Future Work14:53
Summary14:56
Future Work15:05
Thank You!15:06