robert@cs.strath.ac.uk (Robert B Lambert) (03/11/91)
I recently came across an article in Scientific American, Vol.261, July 1989,
which presented an alternative to the Hebb rule.
The article was, "Memory Storage and Neural Systems", by Daniel L. Alkon. It
suggested that learning is performed by associating two simultaneous input
events, independent of the cell state. That is the cell does not have to fire
in order for learning to occur.
In the case of Pavlovian learning, an unconditioned stimulus such as the smell
of meat when presented together with a conditional stimulus such as a bell,
will cause a local interaction between the synaptic connections. This inter-
action will result in a strengthening of the conditional synaptic connection.
At the end of the article the author states that a simulation using this "Local
interaction" model was developed, DYSTAL (DYnamically STable Associative
Learning), for pattern recognition. The author then goes on to state that this
simulation was an improvement over many conventional networks, offering improved
learning times, and most important, the ability to accommodate increasing
numbers of elements without a prohibitive increase in computing power.
I have been unable to find any follow-up to this article. If anyone knows of any
articles which present this work in greater detail, I would be grateful if
you could e-mail me at - robert@cs.strath.ac.uk
- Robert.robert@cs.strath.ac.uk (Robert B Lambert) (03/14/91)
Follow-up article for Alternative to Hebb learning rule. Thanks to those of you who gave me details on the follow up to Alkon's article in Scientific American, July 1989. I have had a large number of requests for additional information, so here is the reference to the article that covers the structure of the DYSTAL network. D. Alkon et al., Pattern Recognition by an Artificial Network Derived from Biologic Neuronal Systems, Biological Cybernetics, Vol. 62, pp 363 - 376, 1990. The article is well worth reading. I would be interested to hear some opinions, so please e-mail me. - Robert. Robert B Lambert E-Mail : robert@cs.strath.ac.uk Dept. Computer Science University of Strathclyde Scotland.