Training free counterpropagation networks as static hetero-associative memories

Abstract

<span style="font-size:11.0pt;mso-bidi-font-size:10.0pt; line-height:115%;font-family:" calibri","sans-serif";mso-ascii-theme-font:minor-latin;="" mso-fareast-font-family:"times="" new="" roman";mso-fareast-theme-font:minor-fareast;="" mso-hansi-theme-font:minor-latin;mso-bidi-font-family:"times="" roman";="" mso-ansi-language:en-us;mso-fareast-language:en-us;mso-bidi-language:ar-sa"="">Counter propagation networks (CPN) that belong to the category of self organization networks, function as statistically optimal self programming look up table. However, the weight adjustment criterion between the input and the competitive layer that follows Kohonen’s unsupervised learning rule, and the same between the competition and the interpolation layer which follows Grossberg's supervised learning rule, can be dispensed with, without affecting the learning ability of the network. Such a network, termed as Training free counter propagation network (TFCPN) is computationally efficient. Also, associative memories which are a class of neural networks are content addressable and possess the capability to store a large set of patterns as memories. In this paper, the TFCPN scheme and its behaviour as a static hetero-associative memory model is discussed. The TFCPN model has been successfully applied to an example in pattern classification and in the design of fink truss in the field of structural engineering.</span>