Neural Networks

Document Sample
Neural Networks Powered By Docstoc
					   Neural Networks

Presented By: Stuart Monaghan
        Date: 18/05/02
                   Contents
•   Introduction to Neural Networks (NN)
•   Definition
•   Background
•   History of Neural Networks
•   Types of Neural Networks
•   Who is concerned with Neural Networks?
•   Advantages/Disadvantages
•   Applications
•   Conclusion
                     Introduction
• Why would anyone want a new sort of Computer?
  – What are Computers good at?
      • Fast Arithmetic.
      • Doing precisely what the programmer programs them to do.
  – What are Computers not good at?
      • Interacting with data concerned with the environment.
      • Able to do large parallel computation.
      • Fault tolerance.
      • Adapting to circumstance.
                 Definition
• Neural Networks
  – Also known as Artificial Neural Networks.
  – They consist of many simple processing units
    (or neurons) connected together.
  – Has the ability to learn from the environment,
    and adapt to it in a manner similar to human
    beings.
                Background
• The Analogy to the Brain
  – Modeled after the structure of the brain.
• The Biological Neuron
  – Most basic element of the human brain =
    neuron.
  – Ability to remember, think, and apply previous
    experiences to our every action.
The Human Brain & The Neural Network
                  History
• During the 40’s enthusiasm was high.
• The Manhattan Project
• Was believed that with enough resources
  the problem of intelligence could be solved.
• Disappointment followed.
                        Quote
• “In the 1960 it was predicted that within ten years
  computers would convert ordinary speech and handwriting
  to print, comprehend and compose natural language, drive
  trucks, do housework, and tutor students better than
  teachers could. Thirty years later many proponents see no
  reason to change these predictions; they still expect them
  within ten years”. Thomas K. Landauer
                             Types
• There are many kinds of Neural Networks!
• Two ways of implementing NN.
   – 1 Feedforward
       • the connections between the units do not form cycles.
       • Response to data input is fast.
   – 2 Feedback
       • There are cycles in the connections.
       • NN must loop for a long time before generating a response.
       Feed forward topology




Courtesy of URL:http://cse.stanford.edu/classes/sophomore-college/projects-00/neural-
                      networks/Architecture/feedforward.html
            Who is Concerned?
• Neural Networks involve the following
  people.
     •   Computer Scientists
     •   Statisticians
     •   Engineers
     •   Biologists
     •   Philosophers
     •   Amongst others...
       Advantages/Disadvantages
• Advantages
      • Start processing data without any preconceived hypothesis.
      • Unbiased and better understanding of data.
      • Once trained, speed and accuracy increases.
      • In return cost is saved.

• Disadvantages
       • Training the network needs experts and is time consuming
       • Does not give reasons for decisions.
                Applications
• Neural Networks can be used for
  forecasting applications such as:
     • Investment Analysis - stock currencies prediction
     • Signature Analysis
               Conclusion
• The subject of Neurocomputing is vast and
  constantly expanding into something that
  can change the way computers operate.
• I think Neural Networks has the potential to
  grow into changing our perception of
  Artificial Intelligence.

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:7
posted:8/25/2012
language:English
pages:14