Hopfield Neural Networks and Physical Systems英文电子书.pdf

Proc. NatL Acad. Sci. USA Vol. 79, pp. 2554-2558, April 1982 Biophysics Neural networks and physical systems with emergent collective computational abilities (associative memory/parallel processing/categorization/content-addressable memory/fail-soft devices) J. J. HOPFIELD Division of Chemistry and Biology, California Institute of Technology, Pasadena, California 91125; and Bell Laboratories, Murray Hill, New Jersey 07974 Contributed by JohnJ. Hopfweld, January 15, 1982 ABSTRACT Computational properties of use to biological or- calized content-addressable memory or categorizer using ex- ganisms or to the construction of computers can emerge as col- tensive asynchronous parallel processing. lective properties of systems -having a large number of simple The general content-addressable memory of a physical equivalent components (orneurons). Thephysical meaning ofcon- tent-addressable memory is described by an appropriate phase system space flow of the state of a system. A model of such a system is Suppose that an item stored in memory is H. A. Kramers given, based on aspects ofneurobiology but readily adapted to in- G. H. Wannier Phys. Rev. 60, 252 (1941). A general content- tegrated circuits. The collective properties of this model produce addressable memory would be capable of retrieving this entire a content-addressable memory which correctly yields an entire memory item on the basis of sufficient partial information. The memory from any subpart of sufficient size. The algorithm for the input Wannier, (1941) might suffice. An ideal memory time evolution of the state of the system is based on asynchronous could deal with errors and retrieve this reference even from the parallel processing. Additional emergent collective properties in- input Vannier, (1941). In computers, only