2: Understanding the Bias with Python: Analyzed and Explained with a Practical Sense (Tutorial Series: Programming Artificial Neural Networks Step by Step with Python) by Eric Joel Barragán González

Author:Eric Joel Barragán González [Barragán González, Eric Joel]
Language: eng
Format: epub
Publisher: UNKNOWN
Published: 2017-03-22T07:00:00+00:00

A 1 0.01 0.75 0.75 1.51

B 1 0.01 0.75 0.75 1.51

A 2 1.01 0.75 0.75 2.51

B 2 0.01 -0.25 -0.25 -0.49

A 3 1.01 -0.25 -0.25 0.51

B 3 1.01 -0.25 -0.25 0.51

A 4 2.01 -0.25 -0.25 1.51

B 4 2.01 -0.25 -0.25 1.51

A 5 2.01 -0.25 -0.25 1.51

B 5 2.01 -0.25 -0.25 1.51

The previous one is a simple example, but the same and more complicated can happen, while weaving a complicated network of adjustments so that the Perceptron interpolate the Weights, while each combination competes with some others, with which it shares common Inputs in "1 ", but with different direction of adjustment required, with respect to the value that has at the moment; where Bias may become the main point of conflict to reach the values of convergence in the Weights.

Because of the above there is who prefers to dispense with the Bias and solves with the way in which it encodes the Inputs. For example if we work with a binary system we could move the representation one unit up and not work with the combination of Inputs all in "0", which would equal the value of "0", and then we go through the representation, we would only lose the last one value to represent, but perhaps it does not affect us if we do not occupy it, or if it is, we could easily cover it by adding a further Input to the neuron. Although I comment that I do not agree with this practice, as I explain later the Bias brings much more than can take away making it require more iterations to learn.

Now why can Bias help and speed up learning? After the above, it will be easier for you to understand, that just as there may be combinations that compete for adjustments in the settings of some particular Weights, the opposite may also happen and that the adjustments have the same direction, and that in a single Iteration will be made several adjustments, of different combinations, on the same Weights, of the Inputs in "1" in common, which will accelerate the learning.

Table of Contents

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.