Date: 17 Nov 2006

Recurrent neural network approach for partitioning irregular graphs

* Final gross prices may vary according to local VAT.

Get Access

Abstract

This paper is concerned with utilizing a neural network approach to solve the k-way partitioning problem. The k-way partitioning is modeled as a constraint satisfaction problem with linear inequalities and binary variables. A new recurrent neural network architecture is proposed for k-way partitioning. This network is based on an energy function that controls the competition between the partition's external cost and the penalty function. This method is implemented and compared to other global search techniques such as simulated annealing and genetic algorithms. It is shown that it converges better than these techniques.