Learning Behavior Using Mult iresolution Recurrent Neural Network
We propose the multiresolution recurrekt neural network to learn behavior b d on view and action. Recurrent neural network structure has the multiresolution channel to establish between the view and the action. It is difficult to learn action using only a image generally. We solve this problem by using the 3 kinds of image on the frequency. We control the rnultiresolution vision using Genetic Algorithm. The action sequences is acquired by the pan-tilt camera.
KeywordsGenetic Algorithm Hide Layer Action Sequence Recurrent Neural Network Topology Change
Unable to display preview. Download preview PDF.
- 1.L. Steels,: The Artificial Life Roots of Artifitial Intelligence, Journal of Artificial Life, Vol. 1, No. 1/2, pp. 75–110, MIT Press (1994)Google Scholar
- 3.J. J. Gibson,: The ecological approach to visual perception, Roughton MMin, (1979)Google Scholar
- 6.R. M. Downs and D. Stea,: Maps in mind, New York: Harper & Row, (1977)Google Scholar
- 7.D. E. Rumelhart, G. E. Hinton and R. J. Williams,: Learning internal represcntations by error propagation, in eds D.E. Rumchart and J.L. McClelland, Parallel Distributed Processing, MIT Press, pp. 318–362 (1986)Google Scholar
- 9.A. P. Witkin,: Scale-space filtering, Int. Joint Conf. Art. Intelligence, pp. 1019–1021 (1983).Google Scholar
- 10.R. Thorn,: Structural Stablity and Morphogenesis, W. A. Benjamin, Inc. (1976)Google Scholar
- 11.D. Marr,: Vision, W.FI.Freeeman and Company, (1982)Google Scholar