Outline Why model neural networks? Modeling Neural Networks A - - PDF document

Modeling Neural Networks

Paul Nuytten CPSC 607

Outline

Why model neural networks? A brief look at the neuron. A look at some current works. Adding an evolutionary strategy.

Why Model Neural Networks?

The nervous system is a very complex system with many

hidden properties.

Many experiments cannot be performed in vivo without

destroying the specimen; leaving many questions unanswered.

When looking at a model many of the redundant

structures and processes can be removed leaving a more focused picture.

The nervous system is a very efficient and massively

parallel computational device. Models may capture this property to solve a certain class of problems.

The Neuron The Neuron

The neuron shares many of the same

components as many other types of cells.

It’s the unique structures that make the

neuron a powerful communication and computational device.

The Neuron

The unique

components are

The dendrites The axon The synapses

The Dendrites

The dendrites are short strands that

protrude from the cell body or the soma.

The dendrites are very receptive to

connections from other neurons.

The dendrites carry signals from the

synapses to the soma.

The Axon

The axon is a long

extension from the soma.

Each neuron only has

• ne axon.

The axon is

myelinated if it is insulated with Schwann cells.

The Axon

If the axon is

myelinated the action potential will travel much faster.

The axon carries

action potentials from the soma to the synapses.

The Synapse

The synapses are the connections made

by an axon to another neuron.

When an action potential arrives at a

synapse from the postsynaptic cell, neurotransmitter is released into the synaptic cleft.

The Synapse

The neurotransmitter will interact with ion

channels on the membrane of the postsynaptic cell causing them to open letting some ions into the cell while letting other ions escape.

A synapse is call excitatory if it raises the local

membrane potential of the post synaptic cell.

Inhibitory if the potential is lowered.

What level should be modeled?

Some Current Works

NEURON GENISIS Neural Swarm Evolutionary Artificial Neural Networks

NEURON

NEURON is a simulation environment for

neurons and neural networks.

The NEURON simulation allows a user to

focus on the biological and biophysical aspects of a neurological system.

Great tool for biologists.

NEURON GENISIS

GEneral NEural SImulation System Provides a simulation environment for

biologically realistic models.

Very similar to NEURON. Allows for parallel processing.

Neural Swarm

Uses the concepts of swarm intelligence when

creating neurological models.

Still in its infancy. Instead of designing the network and defining

processes in mathematical terms, the network and processes are allowed to emerge from the simple interactions within the system.

The results are then compared to biologically

• bserved results.

Neural Swarm

Evolutionary Artificial Neural Networks

Focuses more on the computational aspect of

artificial neural networks (using them to solve problems).

Uses outgrowth and pruning rules to grow a

neural network.

Spontaneous neural activity also contributes to

the development of the network.

What is nice about this simulation is its ability to

apply a genetic algorithm to the above rules and the network’s morphology to specialize the network to a given problem.

Evolutionary Artificial Neural Networks Evolutionary Artificial Neural Networks

Conclusion

When creating a simulation it is important

to identify the level at which to model.

It is also important to identify the target

audience and intended use of your simulation.

Start simple and gradually add complexity. Collaborate.

References

• F. Bloom, C. Nelson, A. Lazerson. Brain, Mind and Behavior Third
• Edition. Worth Publishers, USA, 2001.
• N. Campbell. Biology Fourth Edition, pages 993-1009. The

Benjamins/Cummings Publishing Company, Inc., Menlo Park, California, 1996.

• R. Rojas. Neural Networks A Systematic Introduction. Springer-

Verlag, Berlin, 1996.

• Rust A.G., Adams R., Schilstra M. and Bolouri H. Evolving

computational neural systems using synthetic developmental

• mechanisms. 2003.
• http://www.rwc.uc.edu/koehler/biophys/4d.html
• http://www.neuron.yale.edu/neuron/
• http://www.genesis-sim.org/GENESIS/
• http://strc.herts.ac.uk/bio/alistairr/neural_interests.html