The Improvement of Back Propagation of Neural Networks

الملخص

This paper is concerned with the study of BP network and its evolvement as it is noticed that problems that exist in the Error back propagation algorithm are slow convergence and unlikely to approach minimum. The problems mentioned above led us to find improvements to accelerate this algorithm. In this research BP networks have been constructed, which includes accelerating procedures by using one of the gradient methods to get rid of zigzag phenomena and to approach global minimum. An adaptive learning rate of two methods has been suggested to reduce errors and accelerate training. it has been noticed that calculating appropriate initial weights (Suggested initial, Nguyen–Widrow) and using drive logarithmic function and an adaptive learning rate method have been suggested to reduce errors and accelerate training leading to adequate results and fast learning. A standard network has been applied to all suggestions to diagnose OSTEOARTHRITIN then, results of improving methods which we use have been displayed.