4.9. Choice of the Network Size
In the previous sections, we assumed the number of layers and neurons for each layer to be known and fixed. How one determines the appropriate number of layers and neurons was not of interest to us. This task will become our major focus now.
One answer to the problem could be to choose the size of the network large enough and leave the training to decide about the weights. A little thought reveals that such an approach is rather naive. Besides the associated computational complexity problems, there is a major reason why the size of the network should be kept as small as possible. This is imposed by the generalization capabilities that the network must possess. As has already been pointed out in Section 3.7
Get Pattern Recognition, 4th Edition now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
