Mr. Data from Star Trek          Shape shifting T1000 from Terminator 2    

 

 

Artificial Neural Networks

 

 

·         An artificial neural network is a new type of CPU modeled after the structure of the human brain.

 

·         Modern day computers can perform routine tasks such as complex mathematics very well.  However, a computer that would be based off a neural network could actually “learn” from it experiences, and adapt what it has learned to new and more complex problems.

 

·         Artificial neural networks could pave the way for advances in the field of artificial intelligence, which could lead to the development of so called “learning machines.”

 

 

In order to have a good understanding of how an artificial neural network could create a machine that is capable of learning, it is important to first understand how organic neural networks function.  The importance of understanding organic intelligence stems from the fact that non-organic brains use organic minds as a blueprint.

 

An organic neuron     

 

¨       The majority of the human brain is made up of billions of cells called neurons that do not regenerate.

 

¨       Neurons can connect with up to 200,000 other neurons (1,000-10,000 is typical), thus forming a “neural network.”

 

¨       Each neuron receives input from some source, combines inputs to perform an operation, and finally outputs the operation.  For example, you taste a lemon (the input), millions of neurons in your brain transform that taste into electrical signals (the operation), finally                                                     you feel the sensation of sourness (the output).

 

 

Since the human brain is highly sophisticated, the goals of researchers working on current artificial neural networks is simply to have computers that are capable of solving more complex and learning based problems.  Artificial neural networks will use a simplified version of the neurons present in human brains to create a machine that is capable of learning.

                       

¨       Just as in organic neurons, artificial neurons will contain input, processing, and output areas.

§         Some input will enter the artificial neuron (usually by a sensor). 

§         The input will be multiplied by a weight, which will determine the input’s importance. 

§         These weights are then summed and processed through what is called a “transfer function.” 

§         Finally, the processed input will then be converted into some output.

 

¨      
Artificial neural networks get their power from the fact that billions of such artificial neurons are combined into miniature networks.  Such complex networks of simple components could give future computers the ability to think and reason on their own.

 

Above is an illustration of an artificial neuron.

 

 

Today there exists two different ways in which an artificial neural network can be taught.

 

·         Supervised- This teaching method gives the neural network both the inputs and outputs for a desired problem.  It is then the network’s responsibility to process the inputs and compare the results to the given outputs.  Any errors that occur in the processing of the inputs are refined until the correct outputs are achieved. When the correct outputs are achieved, the neural network is said to have “learned the procedure for finding the correct outputs.”

 

·         Unsupervised- Unlike supervised teaching, unsupervised teaching gives the neural network the inputs but not the desired outputs.  It is up to the network to decide on its own how to solve the problem and discover the correct outputs.  Currently, this teaching technique is not understood very well, but some day it could lead to the type of intelligent robots seen in science fiction movies. 

 

 

Most futurists believe that neural networks will have an increasing role in the development of more life-like robots called androids, which will have the capacity to think and reason just like a human being.  In addition, researchers are working on advanced microprocessors, which use neural networks to incorporate what is known as “fuzzy logic” into them.  Fuzzy logic will give computers the ability to distinguish between the gray areas of life, such as is a person really tall, really short, or kind of tall.

 

 

As current microprocessors begin to reach their physical limitations of speed and capabilities, computer engineers are looking for new and innovative ways to build faster and more capable processors.  One way to approach this problem may be with artificial neural networks, which could give computers the ability to not only perform simple and routine tasks, but actually learn from their interactions with human users and the rest of the world.  Computer-based neural networks will be an ever growing and exciting field of the future, and one that should be watched closely.

 

 

Artificial Neural Network Technology, A DACS State-of-the-Art Report, Dave Anderson and George McNeill, © 1992. Kaman Sciences Corporation.

 

 

 

Peter Lloyd

QMCS 425

March 21, 2002