Using hybrid techniques for a continuous-time inference network

A class of binary relation inference network was recently proposed for constraint-satisfaction applications, including time/location-referencing and air-traffic control problems. There are some intrinsic weaknesses underlying this type of discrete-time inference network, namely, network instability and oscillation under specific circumstances, and the slow convergence rate commonly observed in large networks. To circumvent the potential shortcomings of existing inference networks, state-space techniques are used to derive a more robust continuous-time inference network. Simulation studies on two hybrid implementations indicate significant improvements over discrete-time inference network, and demonstrate their practical viability for applications in time-varying cases