Robust machine learning of the complex-valued neurons

Abstract

Last two decades have witnessed tremendous work in the field of Neurocomputing. A neural network works like a distributed processor which runs parallel. A neural network consists of processing units and the attributes of these units are to acquire knowledge by the virtue of a training process and storing this experimental knowledge in synaptic weights so that it is available for future use. In the present scenario the researchers are developing Artificial Neural Networks (Single / Multilayer) for solving multidimensional problems such as pattern recognition, prediction, optimization, associative memory, control and classifier since these are identified as robust tool various applications. The conventional parameters of Neural Network are generally real numbers and these parameters are only capable to deal with real valued data. But, according to the latest scenario in the field, there is requirement of analysis of high-dimensional data. Therefore, high-dimensional neural networks like CVNN came into the existence. Due to their diversity and abundance it is now becoming difficult to represent Neural Networks in complex domain consequently they started facing representational problems.