4.6. The Backpropagation Algorithm
The other direction we will follow to design a multilayer perceptron is to fix the architecture and compute its synaptic parameters so as to minimize an appropriate cost function of its output. This is by far the most popular approach, which not only overcomes the drawback of the resulting large networks of the previous section but also makes these networks powerful tools for a number of other applications, beyond pattern recognition. However, such an approach is soon confronted with a serious difficulty. This is the discontinuity of the step (activation) function, prohibiting differentiation with respect to the unknown parameters (synaptic weights). Differentiation enters into the scene as a result of the ...
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.
