Reconsidering Cegar: learning good abstractions without refinement | IEEE Conference Publication | IEEE Xplore