Masahiko Morita, Shuji Yoshizawa and Kaoru Nakano
Autocorrelation associative memory is one of the most basic neural
network models for memory, but its dynamical behavior in the
recalling process is not understood well.
The present study deals with the dynamics of the associative
memory and presents a new method called ``partial reverse method''
to improve the recalling process.
Using this method, one can raise both recollection ability and
memory capacity of the network without changing its connections
or learning rules.
The results throw light on the dynamical structure of
associative neural networks, and the partial reverse method may
possibly be applied to some optimization problems.