Quantum-Mechanical Model of Neural Network Elements

被引：0

作者：

Novoselov, A. A. ^{[1
]}

Pavlovsky, O. V. ^{[1
,2
]}

机构：

[1] Moscow MV Lomonosov State Univ, Fac Phys, Moscow 119991, Russia

[2] Kurchatov Inst, Natl Res Ctr, Alikhanov Inst Theoret & Expt Phys, Moscow 117218, Russia

来源：

PHYSICS OF ATOMIC NUCLEI | 2020年 / 83卷 / 12期

基金：

俄罗斯科学基金会;

关键词：

artificial neural networks; path integral; PIMC;

D O I：

10.1134/S106377882011006X

中图分类号：

O57 [原子核物理学、高能物理学];

学科分类号：

070202 ;

摘要：

A model of a neural-network-like quantum-mechanical system is considered. The role of neurons is played by quantum-mechanical particles which evolve under the action of an external double-well potential. An instanton is used as a spike. Neurons are connected by the interaction potential V-int((q) over cap (i), (q) over cap (j)). The connection is directional owing to asymmetry of the interaction potential. Therefore, the complete Hamiltonian of the system is (H) over cap = Sigma(i)(1/2 (p) over cap (2)(i) +V-0((q) over cap (i)) + Sigma(i>j) V-int((q) over cap (i), (q) over cap (j)). The system is investigated by the path integral Monte Carlo (PIMC) method. It is shown that with a certain choice of parameters in this system it is possible to transfer activity in long chains of neurons.

引用

页码：1667 / 1670

页数：4