Efficient VLSI Implementation Based On Constructive Neural Network Algorithms
Iosr Journals, K. R. Ramkumar, S. Sivasankari · Figshare · 2015
C-Mantec is a new neural network algorithm that adds competition between neurons with thermal perceptron learning rule. The neuron learning is ruled by the thermal perceptron rule that guarantees the stability of the learnt information while the architecture increases and while the neurons involve in competition for newly incoming data. Competing makes it possible that even after new neurons have been added to the architecture, existing neurons can still learn if the incoming data is similar to that of the stored information, and this sets up a major difference with the existing constructive neural network algorithms. The new algorithm is tested with diabetes, a well-studied and needed real world problem. It has been used to analyze the size of the neural architecture and the generalization ability in order to compare it with those of other neural classification algorithms. The problem of underfitting/overfitting is also analyzed, and a new method to remove noisy samples is also created. Therefore, it has been shown that the new algorithm generates very small neural architectures with very good generalization capabilities.