Classical and Brain-inspired Neural Networks for Substance Identification and Breast Cancer Detection: The Chaos Challenge

Artificial Neural Networks (ANNs) are largely used in multiple domains and represent an offshoot of pattern recognition paradigms. This work studies the use of three different ANN models, applied as classification technics for breast cancer identification, and as an intelligent data processing module for a multi-wavelength optoelectronic biochemical sensor. The models developed are such as; Multilayer Neural Network (MLNN) trained with the resilient Backprop, a heteroassociative Bidirectional Associative Memory (BAM) and a Chaotic-BAM (CBAM). The later model is tested on 23 different chaotic output functions; the memory fully satisfies the exigencies of a perfect recognition with one specific chaotic map. This work is assessing classification accuracy and investigates the facts that, beyond its biological plausibility, chaos is potential to enhance classification accuracies even in high dimensional application domain. It is resilient to noise present in data; moreover, chaos can face multi class problems with a better efficiency comparing with other classical ANN models. To assess the cited facts, an experimental set up was realized leading to interesting conclusions.