Conference paper
Continuous adaptive critic designs
Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., Vol.5, pp.3001-3006
IEEE
International Joint Conference on Neural Networks, IJCNN 2005 (Montreal, Canada, 31/07/2005–04/08/2005)
2005
Abstract
A continuous formulation of an adaptive critic design (ACD) is investigated. Connections to the discrete case are made, where backpropagation through time (BPTT) and realtime recurrent learning (RTRL) are prevalent. A second order actor adaptation, based on Newton's method, is established for fast actor convergence. Also, a fast critic update for concurrent actor-critic training is outlined that keeps the Bellman optimality correct to first order approximation after actor changes.
Details
- Title
- Continuous adaptive critic designs
- Authors/Creators
- T. Hanselmann (Author/Creator)L. Noakes (Author/Creator)A. Zaknich (Author/Creator)
- Publication Details
- Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., Vol.5, pp.3001-3006
- Conference
- International Joint Conference on Neural Networks, IJCNN 2005 (Montreal, Canada, 31/07/2005–04/08/2005)
- Publisher
- IEEE
- Identifiers
- 991005542100007891
- Copyright
- © 2005 IEEE
- Murdoch Affiliation
- School of Engineering
- Language
- English
- Resource Type
- Conference paper
- Note
- In Proceedings of the International Joint Conference on Neural Networks, 2005. IJCNN '05, Pages 3001-3006.
Metrics
244 File views/ downloads
161 Record Views