Optimization Algorithms for the Minimum-Cost Satisfiability …成本最小的可满足性算法….pptVIP

* * * * * * * * * * * * * * * * Local Search Initialization ex5 16.3 15.1 exam.pi 20.8 5.4 max1024 143.9 27.5 … random initialization of local search is not effective, see the table of runtimes (in cpu-seconds): rounded LP solution random benchmark 4.7 300** bench1.pi **timeout value * Factor 3: Tree-Exploration Strategy exam.pi 7.0 5.4 bench1.pi 7.9 4.7 ex5 16.2 15.1 … search using priority queue performs better than depth-first search (recursion - the usual method): priority* depth-first benchmark 27.5 74.1 max1024 * smallest lower bound first (to try to reduce UB quickly) (runtimes in cpu-seconds) * Experimental Results The solvers Scherzo [Coudert, DAC1996] Cplex (State of the art IP/ILP Solver) Eclipse-lpr (LP Relaxation) Eclipse-cp (Cutting Planes) The benchmarks Logic minimization (unate covering) FSM minimization (binate covering) * Unate Results (nodes visited) (timed out) “harder” * Typical Results: Unate (runtime) timeout * Typical Results: small binate eclipse-cp spends a lot of time with cuts at each node (cplex visits more nodes and does cuts less frequently) * Large binate benchmarks ... but spend more time We visit fewer nodes... CPLEX only does cuts at some nodes. We do cuts everywhere -- could be more judicious * latte = espresso - mincov + eclipse-cp ex5 --- 12** 129.5 139.2 total cover total cover benchmark espresso (hours) latte (secs) max1024 --- 12** 329.1 329.5 14.7 12.9 12** --- prom2 **timeout ex5 256 74 65 max1024 1024 274 259 prod-latte prod-heur prod-orig benchmark 287 287 287 prom2 * The Last Words … Acknowledgments: ? for the benchmarks ? for the solvers (aura, bsolo, espresso, scherzo) ? for constructive reviews Under construction: ? code tune-ups before the release of eclipse ? additional experiments -- for the journal submission -- for comprehensive web-posting of results * Our paper addresses the unate covering and binate covering problems. They are