Can I read Gradient Descent on EtoBox?

Gradient Descent by Christopher M. Bishop; Hugh Bishop is a book available to read on EtoBox.

What is Gradient Descent about?

In the previous chapter we saw that neural networks are a very broad and flexible class of functions and are able in principle to approximate any desired function to arbitrarily high accuracy given a sufficiently large number of hidden units. Moreover, we saw that deep neural networks can encode inductive biases corresponding to hierarchical representations, which prove valuable in a wide range of practical applications. We now turn to the task of finding a suitable setting for the network parameters (weights and biases), based on a set of training data. As with the regression and classification models discussed in earlier chapters, we choose the model parameters by optimizing an error function. We have seen how to define a suitable error function for a particular application by using maximum likelihood. Although in principle the error function could be minimized numerically Section 6.4 through a series of direct error function evaluations, this turns out to be very inefficient. Instead, we turn to another core concept that is used in deep learning, which 209

Author

Christopher M. Bishop; Hugh Bishop

Publisher

Springer International Publishing AG

Published

2023

Language

EN

ISBN

9783031454677

Subjects

Management, Computer Science, Mathematics