English

Gauged Neural Network: Phase Structure, Learning, and Associative Memory

Disordered Systems and Neural Networks 2009-11-07 v4 High Energy Physics - Lattice q-bio

Abstract

A gauge model of neural network is introduced, which resembles the Z(2) Higgs lattice gauge theory of high-energy physics. It contains a neuron variable Sx=±1S_x = \pm 1 on each site xx of a 3D lattice and a synaptic-connection variable Jxμ=±1J_{x\mu} = \pm 1 on each link (x,x+μ^)(μ=1,2,3)(x,x+\hat{\mu}) (\mu=1,2,3). The model is regarded as a generalization of the Hopfield model of associative memory to a model of learning by converting the synaptic weight between xx and x+μ^x+\hat{\mu} to a dynamical Z(2) gauge variable JxμJ_{x\mu}. The local Z(2) gauge symmetry is inherited from the Hopfield model and assures us the locality of time evolutions of SxS_x and JxμJ_{x\mu} and a generalized Hebbian learning rule. At finite "temperatures", numerical simulations show that the model exhibits the Higgs, confinement, and Coulomb phases. We simulate dynamical processes of learning a pattern of SxS_x and recalling it, and classify the parameter space according to the performance. At some parameter regions, stable column-layer structures in signal propagations are spontaneously generated. Mutual interactions between SxS_x and JxμJ_{x\mu} induce partial memory loss as expected.

Keywords

Cite

@article{arxiv.cond-mat/0203136,
  title  = {Gauged Neural Network: Phase Structure, Learning, and Associative Memory},
  author = {Motohiro Kemuriyama and Tetsuo Matsui and Kazuhiko Sakakibara},
  journal= {arXiv preprint arXiv:cond-mat/0203136},
  year   = {2009}
}

Comments

17 pages, 18 figures. Final Version