Optimal task assignment using a neural network

Abstract

Summary form only given. A neural network is described that solves the problem of optimally assigning tasks to processors in a message-passing parallel machine. This task assignment problem (TAP) is defined by creating a task assignment cost function that expresses the cost of communication overhead and load imbalance. TAP is a kind of combinatorial optimization problem which can be solved efficiently by using a neural network, but the Hopfield and Tank approach has certain limitations. The authors have solved these two problems by use of an improved Hopfield model network. By representing TAP in a more direct manner in the neural network, the need for constraints is eliminated, a valid solution is guaranteed, and the number of neurons and connections needed is reduced substantially. >