Home

Contents

Optimization

Neural Nets

  

Finding the Optimum Using Neural Networks

 

Using artificial neural networks doesn't mean you get brainwashed.

----------------------------------------------------------------------------

Index of Page Topics

A Pattern Recognition Tool

Pattern Recognition

How Do Nets Optimize?

Computer Vision

Pattern Recognition

Software

Optimization

Computers

Problem Solving

Computer Games

Neural Networks

Optimization

---------------------------------------------

A Pattern Recognition Tool 

Like the human brain, neural networks are good at pattern recognition. But they aren't as smart or as flexible as even the simplest brain, and they can't yet be conscious or have any level of awareness or understanding. I admit it's another question whether at any level of complexity machines can ever have know-how like the brain, particularly the human variety. On that score I'm not so sure. Depends on what we mean by a machine and what material might be used to construct it.

Ultimately it likely depends on what we find to be the true nature of mind and the true nature of matter itself, the "stuff" we use to build nets. There can be no doubt that they are intimately connected, but we haven't come to the end of the road on either. Even so, doubt doesn't stop us from using the network as an optimization tool, even when dealing with human behavior. The pertinent question then is "What does it mean to use a neural network as a tool for optimization?"

Back to Index

 

--------------------------------------------- 

How Do Nets Optimize?

Neural networks come to a decision by way of adjusting the weights, or strengths, of the connections between pairs of its neural elements -- i.e., the influence exerted in the interaction. Determining the weights has been likened (by Cal Tech physicist John Hopfield) to finding a least energy organization for them. It's possible, though, to have many minimum "energy" points in an environment, like the various low points of valleys in a sea of hills and valleys. Each little valley is a so-called local minimum. One set of weights for a best fit may lead to one low point. But there may be other, lower, points in the overall environment. This creates difficulties. However, by "shuffling the deck," as one might say, another set of weights could lead to a minimum that gets closer to the absolute minimum. Jiggling many times can get you closer to the absolute minimum by giving you a larger number of sample points for comparison.

Back to Index

------------------------------------------------

Top of Page