Towards a More Eecient Stochastic Constraint Solver

E-GENET shows certain success on extending GENET for non-binary CSP's. However, the generic constraint representation scheme of E-GENET induces the problem of storing too many penalty values in constraint nodes and the min-connicts heuristic is not eecient enough on some problems. To overcome these two weaknesses and further improve the performance, we propose several modiications. All of them together can boost the eeciency of E-GENET without resorting to modifying the underlying network model or the convergence procedure in an ad hoc manner. The performance of modiied E-GENET also compares well against that of CHIP.