Feature Construction for Back-Propagation

T h e ease of learning concepts f r o m examples in empirical machine learning depends on the attributes used for describing the training d a t a . We show t h a t decision-tree based feature construction can be used to improve the performance of back-propagation ( B P ) , an artificial neural network a l g o r i t h m , b o t h in terms of the convergence speed a n d the number of epochs taken by the BP a l g o r i t h m to converge. We use disjunctive concepts to illustrate feature construction, and describe a measure of feature quality a n d concept difficulty. We show t h a t a reduction in the difficulty of the concepts to be learned by constructing better representations increases the performance of BP considerably. 1 I n t r o d u c t i o n Recent progress in artificial neural networks ( A N N s ) and their use in disparate domains have spurred the interests of researchers in studying various means of i m p r o v i n g t h e m . A N N s have shown promising results for a number...

• Type: Conference
• Year: 1990
• Where: PPSN
• Authors: Selwyn Piramuthu
• Keywords: Artificial Neural Networks, Decision-tree Based Feature, Distributed And Parallel Computing, Feature Construction, PPSN 1990