|
NeuroCOLT
Technical Report NC-TR-99-041
Computing and Learning with Dynamic
Synapses
Wolfgang Maass,
Institute for Theoretical Computer Science
Technische Universität Graz
Anthony M. Zador
The Salk Institute
La Jolla
California
Received:
07-MAY-1999
Abstract
Conventional models of neural
networks assume that synapses are static, i.e., that they change their
``weight'' only on the slow time scale of learning. We discuss in
this report experimental data which show that this assumption is not
justified for biological neural systems. As a matter of fact, this
assumption is also unjustified for all hardware implementations of
artificial neural nets where the sizes of synaptic ``weights'' are
stored by analog techniques.
We review results about the dynamic behaviour of biological synapses,
survey the available
quantitative models for the temporal dynamics of biological synapses,
and discuss possible computational uses of that dynamics. Some consequences
of synaptic dynamics for learning in neural nets are also discussed.
Download Compressed Postscript
|