The Selection of Optimal Neural Network Model Using Probabilistic Expectation

초록

If there are several parameters of neural network model generated from not many training data, the objective criterion is needed to select the most optimal model among the several models. And this criterion must irrelevant to a specific training data. therefore, a objective criterion that can provide any training data with the same result must be needed. In this paper, we propose the method that can minimize the time to train in the case of infinite network considering all case comparing connection weights of neural networks and using probabilistic distribution theory based probabilistic expectation.