Output details
11 - Computer Science and Informatics
University College London
Return to search
Output 0 of 0 in the submission
Output title
A Unifying Perspective of Parametric Policy Search Methods for Markov Decision Processes
Type
E - Conference contribution
DOI
-
Name of conference/published proceedings
Neural Information Processing Systems
Volume number
-
Issue number
-
First page of article
2726
ISSN of proceedings
-
Year of publication
2012
URL
-
Number of additional authors
1
Additional information
<12> Efficiently training Markov Decision Processes is a long-standing and fundamental problem in computer science. In this paper we showed for the first time how to view many existing algorithms can be viewed in a unifying framework, leading us to suggest a novel algorithm. This new algorithm has excellent performance and is arguably the first new practical approach to training MDPs for many years. We trained the world's best Tetris player using this algorithm as an example of its strengths. The paper was an oral at NIPS (around 20 papers of 1500 submitted were accepted as orals).
Interdisciplinary
-
Cross-referral requested
-
Research group
None
Citation count
0
Proposed double-weighted
No
Double-weighted statement
-
Reserve for a double-weighted output
No
Non-English
No
English abstract
-
