Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!uxc!iuvax!rutgers!caip.rutgers.edu!boubez From: boubez@caip.rutgers.edu (Toufic Boubez) Newsgroups: comp.ai.neural-nets Subject: TSP on Hopfield network Keywords: TSP, convergence, parameters Message-ID: Date: 3 May 89 23:23:44 GMT Organization: Rutgers Univ., New Brunswick, N.J. Lines: 23 I am trying to implement a TSP on a Hopfield net, as described in Hopfield and Tank's paper [Computing with Neural Circuits, Science, vol 233, august 86]. I'm having trouble getting it to converge to the right solution (to any solution!). Even when started from the right solution, the net works itself back to a spurious state. I have tried updating the nodes in a non-parallel fashion (it used to oscillate before I did that), using new values from other nodes to update a node's output. I have tried fiddling with the thresholds, the weight values, etc... What I want to know is this: Has anyone tried it successfully, and if so, what are the parameters that you used, and any other hints/suggestions. Thanks a lot. Please e-mail to me. -- ccccccccccccc ccccccccccccccccccc cccc HH HH cccc Toufic Boubez cccc HH HH boubez@elbereth.rutgers.edu cccc HHHHHHH GO HABS GO! boubez@caip.rutgers.edu cccc HH HH cccc HH HH cccc ccccccccccccccccccc ccccccccccccc