LEARNING A SPIN-GLASS - DETERMINING HAMILTONIANS FROM METASTABLE STATES

We study the problem of determining the Hamiltonian of a fully connect ed Ising spin glass of N units from a set of measurements, whose sizes needs to be O(N-2) bits. The student-teacher scenario, used to study learning in feed-forward neural networks, is here extended to spin sys tems with arbitrary couplings. The set of measurements consists of dat a about the local minima of the rugged energy landscape. We compare si mulations and analytical approximations for the resulting learning cur ves obtained by using different algorithms. (C) 1998 Elsevier Science B.V. All rights reserved.