Neural networks with low levels of activity: Ising vs. McCulloch-Pitts neurons
Tóm tắt
The performance of neural networks used as associative memory for uncorrelated patterns with prescribed mean activity is analyzed within the replica symmetric mean field theory. The optimal representation of the possible states of the neutrons, active or inactive, is found to depend on the mean activity. For activity equal one half Ising neurons and for low activities McCulloch-Pitts neurons are optimal. In this optimal representation the noise due to noncondensed patterns is reduced.
Tài liệu tham khảo
Abeles, M.: Studies of brain functions. Berlin, Heidelberg, New York: Springer 1982
Willshaw, D.J., Bunemann, O.P., Longuet-Higgins, H.C.: Nature222, 960 (1969)
Palm, G.: Biol. Cybern.39, 125 (1981)
Tsodyks, M.V., Feigel'mann M.V.: Europhys. Lett.6, 101 (1988)
Buhmann, J., Divko, R., Schulten, K.: (to be published)
Amit, D.J., Gutfreund, H., Sompolinsky, H.: Phys. Rev. A35, 2293 (1987)
Gutfreund, H.: Phys. Rev. A37, 570 (1988)
Parga, N., Virasoro, M.A.: J. Phys. (Paris)47, 1857 (1986)
Feigel'man, M.V., Ioffe, L.B.: Preprint (1986);
Bös, S., Kühn, R., van Hemmen, J.L.: Z. Phys. B — Condensed Matter71, 261 (1988)
Gardner, E.: J. Phys. A21, 271 (1988)
Amit, D.J., Gutfreund, H., Sompolinsky, H.: Ann. Phys. (NY)173, 30 (1987)
McCulloch, W.A., Pitts, W.: Math. Biophys.5 (1943)
Hopfield, J.J.: Proc. Natl. Acad. Sci. USA79, 2554 (1982)
Kohonen, T.: Selforganization and associative memory. Springer 1984. Personnaz, L., Guyon, I., Dreyfus, G.: J. Phys. (Paris)46, 359 (1985)
Horner, H.: Phys. Blätter44, 29 (11088);
Bormann, D., Frick, M., Horner, H., Schmidt, A.: (to be published)
Diederich S., Opper, M.: Phys. Rev. Lett.58, 949 (1987);
Pöppel, G., Krey, U.: Europhys. Lett.4, 979 (1987);
Gardner, E., Stroud, N., Wallace, D.J.: (to be published)
Forrest, B.M.: (to be published)
Rieger, H., Schreckenberg, M., Zittartz, J.: Z. Phys. B — Condensed Matter72, 523 (1988)