Finding Probably Approximate Optimal Solutions by Training to Estimate the Optimal Values of SubproblemsNimrod MegiddoSegev Wasserkruget al.2025arXiv