On the Thermodynamics of the Hopfield Neural Network with External Field
J. Ulaczyk, Robert A. Kosiński, Andrzej Zagorski · Acta Physica Polonica A · 2003
si cs, W arsaw U ni versity of Technolo gy Ko szykowa 75, 00-662 W arsaw, Po l and b Centra l Insti tute of La b or Pro tecti on, Nati ona l R esearch Insti tute Czerni akowska 16, 00-701 Warsaw, Po la nd ( Recei ved A pr i l 17, 2003)In the pap er ther modynam i c prop ertie s of an arti Ùcial neural net w or k are analy zed in a way analo gous to spin glasses theory .Synaptic connections are calcula ted numerically according to the H ebb rule and their distribution is obtained for di˜erent characteristics of stored patterns.T he phase diagrams and magnetizatio n are established in dependence on the temp erature of the netw ork and the external Ùeld (threshold ).It w as show ed that changing control parameters typical of artiÙcial neural netw ork (i.e. numb er of stored patterns and pattern bias level) one obtains the results similar to the Sherrington { K irk patri ck mo del of spin glass.