Authors : C. Vijayalakshmi and k. Senthamarai Kannan
Abstract: This study mainly deals with a general framework for Hidden Markov Models and Neural Networks by using back propagation algorithm. In the training phase an efficient way of updating the weights of neurons based on the relative entopy function is introduced, so that the network converges in a rapid manner. Exponential gradient descent algorithm has been used for updating the weights of neurons in each iteration of the training phase. Here the component of the gradient term appears in the exponent of a factor that is used in updating the weight vector multiplicatively. A scaling factor is derived in which the hidden layer output is linear which represents the total weight on hidden layer nodes. The numbers of hidden layer units were also varied for the various learning rates and the performance were marked. The efficiency and accuracy of learning process greatly depends on the training algorithm used. An algorithm is designed in such a way that a new weight update rule has been introduced. Exponentiated gradient descent weight update rate that substitutes the existing gradient descent update rate of back propagation algorithm in which the results are faster and it is an efficient Markov learning network. The Hidden Neural Network can be viewed as an undirected probabilistic model in which the study of the structure of the standard genetic code is analyzed.
C. Vijayalakshmi and k. Senthamarai Kannan , 2007. Implementation of Artificial Neural Network with Hidden Markov Model for Analysing the Genetic Code. Journal of Engineering and Applied Sciences, 2: 17-29.