Figure shows the physical structure neuron general

Chapter 5. Neural networks

I believe that the techniques for using neural networks efficiently solve many problems that are intractable or difficult using other AI programming techniques. Although most of this book is intended to provide practical advice (with some theoretical background) on using AI
programming techniques, I can not imagine being interested in practical AI programming without also wanting to think about the philosophy and mechanics of how the human mind works. I hope that my readers share this interest and later I will discuss “real neurons”, etc.

• •

We will start this chapter by discussing human neuron cells and what features of real neurons that we will model. Unfortunately, we do not yet understand all of the biochemical processes that occur in neurons, but there are fairly accurate models available (web search “neuron
biochemical”). Neurons are surrounded by thin hair like structures called dendrites, which serve to accept activation from other neurons. Neurons sum up activation from their dendrites and each neuron has a threshold value; if the activation summed over all incoming dendrites exceeds this

5.1 Hopfield neural networks

Hopfield neural networks implement associative (or content addressable) memory. A Hopfield network is trained using a set of patterns. After training, the network can be shown a pattern similar to one of the training inputs and it will hopefully associate the “noisy” pattern with the correct input pattern. Hopfield networks are very different that back propagation networks because the training data only contains input examples. Internally, the operation of Hopfield neural networks is very different that back propagation networks that we will see later in this chapter. We use Hopfield neural networks to introduce the subject of neural nets because they are very easy to simulate with a program, and they can also be very useful in practical applications.