TY - JOUR

T1 - Statistical mechanics of learning via reverberation in bidirectional associative memories

AU - Centonze, Martino Salomone

AU - Kanter, Ido

AU - Barra, Adriano

N1 - Publisher Copyright:
© 2024

PY - 2024/3/1

Y1 - 2024/3/1

N2 - We study bi-directional associative neural networks that, exposed to noisy examples of an extensive number of random archetypes, learn the latter (with or without the presence of a teacher) when the supplied information is enough: in this setting, learning is heteroassociative – involving couples of patterns – and it is achieved by reverberating the information depicted from the examples through the layers of the network. By adapting Guerra's interpolation technique, we provide a full statistical mechanical picture of supervised and unsupervised learning processes (at the replica symmetric level of description) obtaining analytically phase diagrams, thresholds for learning, a picture of the ground-state in plain agreement with Monte Carlo simulations and signal-to-noise outcomes. In the large dataset limit, the Kosko storage prescription as well as its statistical mechanical picture provided by Kurchan, Peliti, and Saber in the eighties is fully recovered. Computational advantages in dealing with information reverberation, rather than storage, are discussed for natural test cases. In particular, we show how this network admits an integral representation in terms of two coupled restricted Boltzmann machines, whose hidden layers are entirely built of by grand-mother neurons, to prove that by coupling solely these grand-mother neurons we can correlate the patterns they are related to: it is thus possible to recover Pavlov's Classical Conditioning by adding just one synapse among the correct grand-mother neurons (hence saving an extensive number of these links for further information storage w.r.t. the classical auto-associative setting). Likewise, it is enough to equip the BAM's synaptic coupling with its transpose matrix (and alternate among these two, e.g., by introducing a clock) to let these networks be able count sequences of patterns (still preserving Detailed Balance, at difference w.r.t. the auto-associative case).

AB - We study bi-directional associative neural networks that, exposed to noisy examples of an extensive number of random archetypes, learn the latter (with or without the presence of a teacher) when the supplied information is enough: in this setting, learning is heteroassociative – involving couples of patterns – and it is achieved by reverberating the information depicted from the examples through the layers of the network. By adapting Guerra's interpolation technique, we provide a full statistical mechanical picture of supervised and unsupervised learning processes (at the replica symmetric level of description) obtaining analytically phase diagrams, thresholds for learning, a picture of the ground-state in plain agreement with Monte Carlo simulations and signal-to-noise outcomes. In the large dataset limit, the Kosko storage prescription as well as its statistical mechanical picture provided by Kurchan, Peliti, and Saber in the eighties is fully recovered. Computational advantages in dealing with information reverberation, rather than storage, are discussed for natural test cases. In particular, we show how this network admits an integral representation in terms of two coupled restricted Boltzmann machines, whose hidden layers are entirely built of by grand-mother neurons, to prove that by coupling solely these grand-mother neurons we can correlate the patterns they are related to: it is thus possible to recover Pavlov's Classical Conditioning by adding just one synapse among the correct grand-mother neurons (hence saving an extensive number of these links for further information storage w.r.t. the classical auto-associative setting). Likewise, it is enough to equip the BAM's synaptic coupling with its transpose matrix (and alternate among these two, e.g., by introducing a clock) to let these networks be able count sequences of patterns (still preserving Detailed Balance, at difference w.r.t. the auto-associative case).

KW - Artificial intelligence

KW - Associative memories

KW - Complex systems

KW - Explainable AI

KW - Machine learning

KW - Statistical mechanics

UR - http://www.scopus.com/inward/record.url?scp=85183453126&partnerID=8YFLogxK

U2 - 10.1016/j.physa.2024.129512

DO - 10.1016/j.physa.2024.129512

M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???

AN - SCOPUS:85183453126

SN - 0378-4371

VL - 637

JO - Physica A: Statistical Mechanics and its Applications

JF - Physica A: Statistical Mechanics and its Applications

M1 - 129512

ER -