AN IMPROVEMENT OF NEURAL NETWORKS APPLIED TO PHARMACEUTICAL PROBLEMS

In applying the neural network to the classification problem in pharma cology, we adopt an extended back-propagation (EBP) learning which adj usts the parameters appearing in an activation function, as well as th e weights. The results of simulations show that such an extended learn ing speeds up the learning process as compared with the conventional b asic back-propagation procedure, irrespective of the initial values of the parameters, which is extremely useful in the practical applicatio n of the neural network in the pharmaceutical field. We have also foun d that use of Morita's activation function beyond the sigmoid type fur ther accelerates the EBP learning in some cases.