Using an Eastern Philosophy for Providing a Theoretical Basis for Some Heuristics Used in Artificial Neural Networks

Abstract

Artificial Neural Networks technology has been recognised as a promising approach to solve a variety of complex problems, which could not be solved otherwise. Despite showing promising results, this technology has often been criticised for the lack of a theoretical basis. This paper presents a theoretical basis for some heuristics used in designing and training of Artificial Neural Networks. Our research exploits Theravada Buddhist theory of mind, which is a well-known Eastern philosophy and also falls under the context of alternative systems of knowledge. The theoretical basis is derived using the concept of thought process in Buddhism. In this sense, firstly, our theory supports the heuristics that the number of neurons in the input layer should be the same as the number of components in an input. The second heuristic, the use of small values as initial weights has been supported. A theoretically based mechanism for initialisations of weights has also been introduced. Further work on the project introduces the novel idea of recursive training of Artificial Neural Networks. Hence, it is concluded that the integration of Western model of Artificial Neural Networks with the Eastern model of thought process in Buddhism will be immensely be beneficial for the development in the field.