Semantic Extraction with the Morphogenetic Neuron (Invited Paper)

In this paper we present a new model of neuron denoted morphogenetic neuron. The name morphogenetic means “form generator”. The form is a non linear map generated from superposition of basis functions. The parameters of the superposition are obtained by cases of the non linear map and by cases of the basis functions. The cases of the basis functions are vectors that form the geometric reference in which we locate the vector of the non linear map cases. The geometric image in the multidimensional space of the non linear functions cases provides the geometric language to describe the new type of neuron. With this new model of neuron we can overcome classical learning problems in the back propagation model of neurons and in the same time we can give a formal description of the population of neurons to code non linear functions. In the cortex every non linear function is represented by a special code of basis functions one for any neuron of the population. In the data mining domain, different methods are invoked for extract the semantic values from the data. For example Reverse Engineering methods transform data in concepts and Case Basis Reasoning uses past cases to solve problems. All the different methods are in many cases separate one from the other. We argue that the morphogenetic neuron can be utilized to give a common framework to include the different data mining methods. Copyright c © 2004-2005 Yang’s Scientific Research Institute, LLC. All rights reserved.