Abstract
This paper characterizes connectionist-type architectures that allow a distributed solution for classes of constraint-satisfaction problems. The main issue addressed is whether there exists a uniform model of computation (where all nodes are indistinguishable) that guarantees convergence to a solution from every initial state of the system, whenever such a solution exists. We show that even for relatively simple constraint networks, such as rings, there is no general solution using a completely uniform, asynchronous, model. However, some restricted topologies like trees can accommodate the uniform, asynchronous, model and a protocol demonstrating this fact is presented. An almost-uniform, asynchronous, networkconsistency protocol is also presented. We show that the algorithms are guaranteed to be selfstabilizing, which makes them suitable for dynamic or error-prone environments.