Associative Memory With Nonmonotone Dynamics
Masahiko Morita
Abstract:
The dynamics of autocorrelation associative memory is examined, and a
novel neural dynamics which greatly enhances the ability of
associative neural networks is presented. This dynamics is such that
the output of some particular neurons is reversed (for a discrete
model) or the output function is not sigmoid but nonmonotonic (for an
analog model). It is also shown by numerical experiments that most of
the problems of the conventional model are overcome by the improved
dynamics. These results are important not only for practical purposes
but also for understanding dynamical properties of associative neural
networks.
Keywords:
Dynamics of associative memory, Associative neural networks,
Autocorrelation associative memory, Recalling process, Memory
capacity, Spurious memory, Partial reverse method, Nonmonotone
dynamics, Memory of correlated patterns.
PostScript manuscript (176KB)
is available