Title :
A dual-phase technique for pruning constructive networks
Author :
Thivierge, J.P. ; Rivest, F. ; Shultz, T.R.
Author_Institution :
Dept. of Psychol., McGill Univ., Montreal, Que., Canada
Abstract :
An algorithm for performing simultaneous growing and pruning of cascade-correlation (CC) neural networks is introduced and tested. The algorithm adds hidden units as in standard CC, and removes unimportant connections by using optimal brain damage (OBD) in both the input and output phases of CC. To this purpose, OBD was adapted to prune weights according to two separate objective functions that are used in CC to train the network, respectively. Application of the new algorithm to two databases of the PROBEN1 benchmarks reveals that this new dual-phase pruning technique is effective in significantly reducing the size of CC networks, while providing a speed-up in learning times and improvements in generalization over novel test sets.
Keywords :
generalisation (artificial intelligence); learning (artificial intelligence); neural nets; PROBEN1 benchmarks; cascade-correlation neural networks; constructive networks pruning; databases; dual-phase pruning technique; input phases; optimal brain damage; output phases; Benchmark testing; Biological neural networks; Computer network management; Computer science; Databases; Network topology; Performance evaluation; Potential well; Psychology; Quality management;
Conference_Titel :
Neural Networks, 2003. Proceedings of the International Joint Conference on
Print_ISBN :
0-7803-7898-9
DOI :
10.1109/IJCNN.2003.1223407
