An Enhanced Linear Programming Approach to the Design and Training of Neural Networks for Pattern Classification

点击翻译按钮获取中文摘要

WPCD 2 B V P Z Courier 10cpi ? x x x , x 6 X @ 8 ; X @ HP LaserJet IIIP HPLSIIIP.PRS x @ , \ , |X @ 2 2 B #| x 9224810 Ignizio Examples of the pattern classification problem (known variously as: pattern recognition, discriminant analysis, and pattern grouping) seem to be found virtually anywhere and everywhere. In general such problems involve the need to assign objects to various groups, or classes. Recent attempts to solve this problem have employed linear programming methods and, in particular, neural networks. In this research an approach will be developed that combines linear programming (specifically, traditional linear programming and/or linear goal programming) with neural networks an approach that should prove to be a significant advance over earlier and ongoing attempts toward the development of such combinations. Ongoing efforts toward neural network design/training via linear programming have failed to either consider or include an adequately rigorous treatment of the specifics of hardware replication. The researchers will develop and evaluate a detailed representation of all pertinent network processing elements and connections in order that direct replication via hardware ultimately be permitted.