BioCAM: Biological Model of Constructing Artificial Neural Networks through Cellular Automata

A. Carrascal, D. Manrique, J. Ríos, and C. Rossi (Spain)


Cellular Automata, Biological Model, Genetic Algorithms, Neural Networks.


A new technique for constructing neural networks is proposed in this paper. This approach, called BioCAM (Biological Cellular Automata Model), takes as its inputs a set of training patterns that define a problem, and gives, in turn, a trained binary recurrent neural network. Our model takes inspiration from neuro-embryology, the biological process of growing of the neural tissues, where cellular automata have been designed to simulate the grown of axons and dendrites that constitute the connections of the artificial neural network that is being formed as final solution. Initial configurations are generated using an evolutionary process that mimics the brain's innate capabilities. Experimental results on laboratory tests and on a real world problem confirm the high performance of BioCAM and demonstrate significant advantages with respect to other automatic neural construction approaches.

Important Links:

Go Back