Solving geometric constraints in a parallel network

The paper describes a network implementation of the SUP-INF method for solving sets of inequalities, giving several advantages over previous implementations. The cost of symbolic manipulation is transferred to compile time allowing an increase in speed at run time due to parallel evaluation. Further, allowing iteration in the network improves the competence of the method when working with nonlinear expressions. The network is used to implement a geometric reasoner for a computer vision program and this is shown to meet the general requirements for such a system.