DocumentCode
2633881
Title
Learning the learning parameters
Author
Pedone, Roberto ; Parisi, Domenico
Author_Institution
Inst. of Psychol., Nat. Res. Council, Rome, Italy
fYear
1991
fDate
18-21 Nov 1991
Firstpage
2033
Abstract
A variation of the backpropagation procedure that dynamically adjusts the values of the learning rate and momentum parameters during learning is proposed. These values are made dependent on the standard deviation of the activation distribution of each hidden unit, which allows the network to adapt the parameter values to each individual weight. The new procedure was applied to a simple categorization task and it gave better convergence results than the standard backpropagation procedure
Keywords
convergence; learning systems; neural nets; activation distribution; backpropagation; convergence; hidden unit; learning parameters; learning rate; learning systems; neural nets; standard deviation; Backpropagation algorithms; Convergence; Councils; Error correction; Frequency; Genetic algorithms; Proposals; Psychology; Shape; Transfer functions;
fLanguage
English
Publisher
ieee
Conference_Titel
Neural Networks, 1991. 1991 IEEE International Joint Conference on
Print_ISBN
0-7803-0227-3
Type
conf
DOI
10.1109/IJCNN.1991.170626
Filename
170626
Link To Document