Abstract
We propose a modified belief propagation algorithm, with over-relaxed dynamics. Such analgorithm turns out to be generally more stable and faster than ordinary beliefpropagation. We characterize the performance of the algorithm, employed as a tool forcombinatorial optimization, on the random satisfiability problem. Moreover, we trace aconnection with a recently proposed double-loop algorithm for minimizing Bethe andKikuchi free energies.
Full Text
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call