Transcript Artificial Neural Systems

Artificial Neural Systems
Intro
• Artificial neural systems try to process information in
the same way as the human brain does.
• Traditional computer systems process data using
the "Von Neuman" model.
• Artificial neural networks (ANNs) try to imitate the
way the human brain is organised and the way the
brain handles information.
Traditional (von neuman) computer
system
• A human written program tells the computer system how
to use inputs and follow a plan to produce appropriate
output.
• The human programmer uses complex theories to make
a sophisticated plan that will make the computer system
successful.
• The instructions in the program are carried out one after
the other at a rate of hundreds of millions of instructions
per second (MIPS) by a single complex and powerful
processor.
• This is NOT how a human brain is physically structured
nor is it the method used by the brain to process
information.
Human brain and neurons
• A human brain has about 200,000 neurons.
• A neuron is the type of brain cell most associated with
intelligent behaviour.
• A neuron is quite a simple "device" but each neuron has
connections to many many other neurons. The pattern of
interconnection is very complex.
• Each neuron receives signals from other neurons which
it may ,(or may not), pass on to other neurons. The brain
processes information by creating complex patterns of
signals (neural pathways) being "fired" around large
groups of neurons.
Human brain and neurons (cont…)
• Any set of signals may be passed at the same time as
other signals are being passed and so the brain operates
with parallel processing, in fact very many process may
take place at the same time. This parallel activity helps
the brain to be fast.
Artificial Neural Network (ANN)
• The ANN tries to imitate the neurons in the human brain.
• The network is composed of a large number of highly
interconnected processing elements (neurons) working
in parallel to solve a specific problem. It imitates the
brain.
• There are several layers of neurons. The ANN has an
input layer, an output layer and one or more hidden
layers in between.
• Data is entered at the input layer and signals are passed
through the connections between layers, though some
signals are stronger than others. Each neuron uses a
calculation based on its input signal to produce the
signal it passes on to the next layer.
• The output layer delivers the final results.
• There is no program!
• The ANN learns how to be successful by training.
• In the first training session, the ANN takes a completely
random guess at the answer (nonsense), getting it wildly
wrong.
• Each training session involves the ANN being given
another problem and the ANN output is compared
against the correct answer. The difference between the
ANN result and the correct answer (error) is fed back
through the ANN.
• The ANN uses the error feedback to alter the pathways
between the layers, some pathways are made stronger,
others weaker. The pathways become tuned to the
correct answer.
• After the ANN has been trained, it will be used to
process new unseen problems of the same type. Usually
the ANN will have a very high success rate at solving the
problem.
• The ANN has learned its own way for solving the
problem.
Applications of ANNs
• Neural networks are best at identifying patterns or trends
in data (pattern matching), so they are well suited to
prediction or forecasting needs including:
• Hand-written word recognition (used for reading
postcodes)
– Have a look some work done at University of
Technology in Sydney Australia
• Stock market prediction; will the shares rise or fall, when
should investors buy or sell?
– Tradescision produces ANN for market analysis, have
a look at their website
• Debt risk assessment; should the bank customer get a loan
or not, what are the chances of not getting (all) the money
back?
• Recognition of speakers (voices) in communications;
• Diagnosis of hepatitis (a liver disease)
• Three-dimensional object recognition (finger print recognition)
• Facial recognition (used by modern digital cameras, police
forces)
– a pattern is some form of sequence or repetition
– eg a person's distinctive voice pattern, a fingerprint, the
pattern of sales for ice cream over 12 months of the year
– ANN is good at any application where there is pattern
identification
– here is a military use of ANN that did not go as planned