Artificial Intelligence: What You Need to Know About Machine Learning, Robotics, Deep Learning, Recommender Systems, Internet of Things, Neural Networks, Reinforcement Learning, and Our Future by Wilkins Neil

Author:Wilkins, Neil [Wilkins, Neil]
Language: eng
Format: azw3
Published: 2019-01-24T16:00:00+00:00

Chapter 7: Neural Networks

If humans are intelligent because of their brains, and if brains work by creating neural connections called synapses, wouldn’t it make sense to simulate these networks of connections in order to simulate intelligence in machines? Or at the very least, that is what early AI researchers thought. The sheer volume of connections in the human brain is what we owe our intelligence to. The average human brain has around one hundred billion neurons or ten to the eleventh power. These neurons can then connect to up to 7,000 other neurons – meaning the total number of connections is an order of millions of billions of connections. That is bananas. The beginnings of artificial neural networks (ANN) directly coincided with the study of real neural networks. In 1943, a neurophysiologist named Warren McCulloch teamed up with mathematician Walter Pitts to describe how neurons in the brain might work. They co-authored a paper in which they created a simple ANN out of electrical circuits. The ANN they designed used artificial or logical neurons called the McCulloch-Pitts neuron.

Inside the brain, a neuron works by receiving inputs, processing the information, and then transmitting it to other neurons. A neuron cell is made out of a nucleus that forms the cell body. From the cell body, structures called dendrites branch out like the arms of an octopus. Attached to the cell body is a long chain-like structure called the axon that is used to connect to other neurons. This point of connection is called the synapse and also looks like tendrils or branches used for connecting. The dendrite structure receives information and the cell body or soma processes it. The output then gets fired through the axon and into the synapse where the next neuron receives it. This is, of course, a simplified version of the true story, but it is enough for understanding artificial neurons and ANNs.

The McCulloch-Pitts neuron is, of course, purely logical. It doesn’t make sense to talk about parts of a neuron as if they existed in real life. But it does make sense to talk about the parts according to their logical function. These artificial neurons are made up of two parts simply called f and g. First is g, it acts as the dendrite and receives some input, performs some processing, and passes it on to f. The processing can be a chain of Boolean operations that are said to be either excitatory or inhibitory decisions. A decision that is inhibitory has a greater effect on the neuron firing or not. For example, if the neuron is deciding whether to eat at a restaurant, an inhibitory decision would be something like “Am I hungry?” Obviously, if you aren’t hungry, you won’t make the trip. Less important decisions in the process may be “Do I crave fast food?”, “Do I feel like going out?”, “Does my car have enough gas?” and so on. These other excitatory inputs will not make the final decision on their own, but together they might.

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.