Using Adaptive Resonance Theory in Design of Structures
Ardavan Avazdavani, Soudabeh Borazjani · 2005
In this research the basic algorithm of ART2 neural network has been modified for proper and efficient classification of vectors. In the basic architecture of ART2, the length of vectors is neglected. This causes error in sorting; parallel vectors with different length are classified in the same category. To overcome this deficiency, a virtual input neuron is added to consider vector length. The modified architecture not only considers the similarity of vectors direction but also considers the magnitude of vectors in sorting. ART neural networks are classified as unsupervised learning nets, a method is presented for supervised learning of ART2 without general changes in the basic algorithm. In this method ART2 net is used to automatic design of structural members. Finally this modified algorithm is used for design of a sample building structure.