Title :
Stochastic bidirectional training
Author_Institution :
Dept. of Inf. Eng., New South Wales Univ., Sydney, NSW, Australia
Abstract :
We consider connectionist compression schemes using auto-associative networks, demonstrate the advantages gained by imposing different constraints on allowed network weights, and give a comparison with pruning of the unconstrained auto-associative network. In this paper we demonstrate the advantages for generalisation performance of constraining weights symmetrically using weight sharing, and by constraining functional symmetry by the use of enhanced backpropagation networks trained bidirectionally. In the process, we derive the stochastic bidirectional training algorithm
Keywords :
backpropagation; feedforward neural nets; generalisation (artificial intelligence); network topology; auto-associative networks; backpropagation networks; bidirectional learning; feedforward neural networks; functional symmetry; generalisation; network weights; pruning; topology; weight sharing; Backpropagation algorithms; Computer science; Costs; Electronic mail; Image coding; Network topology; Neural networks; Neurons; Stochastic processes; Switches;
Conference_Titel :
Systems, Man, and Cybernetics, 1998. 1998 IEEE International Conference on
Conference_Location :
San Diego, CA
Print_ISBN :
0-7803-4778-1
DOI :
10.1109/ICSMC.1998.728185
