Study on analog circuit implementation of ReLU Hopfield Neural Network for inequality-constrained optimization problems
TL;DR
This paper applied a circuit implemented ReLU Hopfield Neural Network to a mathematical problem with an inequality constraint and implemented a circuit and observed converged neural circuit outputs that corresponded well to theoretical results and simulations.
Abstract
Though an eminent performance of Artificial Intelligence (AI) is expected to be applied in various technological fields, its high energy consumption is still an issue to be solved for the sustainable implementation of AI into our society. The analog implemented neural networks are promising alternative computation devices with their high-speed convergence and low energy consumption. In this paper, we applied a circuit implemented ReLU Hopfield Neural Network to a mathematical problem with an inequality constraint. After confirming the correspondence between the system dynamics and the search algorithm, we implemented a circuit and observed converged neural circuit outputs that corresponded well to theoretical results and simulations. The objective function value obtained by the proposed analog circuit was within 1.1% relative error of the optimal value.