Title :
One-shot algorithm for temporal sequences
Author :
Demura, Kosei ; Kajiura, Masahiro ; Anzai, Yuichiro
Author_Institution :
Dept. of Comput. Sci., Keio Univ., Yokohama, Japan
fDate :
27 Jun-2 Jul 1994
Abstract :
Recurrent SOLAR (Supervised One-shot Learning Algorithm for Real number inputs) requires only a single presentation of an analog training set for learning temporal sequences. Recurrent SOLAR does not use the gradient decent algorithm, so it has no local minima problem, no topological problem and extraordinary speed-up compared to the algorithms based on the gradient decent method
Keywords :
feedforward neural nets; learning (artificial intelligence); recurrent neural nets; sequences; analog training set; recurrent SOLAR; supervised one-shot learning algorithm for real number inputs; temporal sequences; Application specific processors; Computer science; Context modeling; Hazards; Large-scale systems; Network topology; Spatiotemporal phenomena; Supervised learning; Symmetric matrices; Tiles;
Conference_Titel :
Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Conference_Location :
Orlando, FL
Print_ISBN :
0-7803-1901-X
DOI :
10.1109/ICNN.1994.374294
