The retrieval possibilities of autoassociative networks under variation of the interconnection architecture

The dependence of the main retrieval characteristics of autoassociative neural networks-the memory capacity and the basins of attractors-on the architecture of connections is studied in the framework of the macrodynamical approach. The continuous-time version of the macrodynamical system of equations is analyzed. It is shown that in the limit of diluted connection structure the retrieval possibilities of the autoassociative network tend to those for a corresponding layered network.<>