sk_p: a neural program corrector for MOOCs
Yewen Pu, Karthik Narasimhan, Armando Solar-Lezama, Regina Barzilay · 2016
We present a novel technique for automatic program correction in MOOCs, capable of fixing both syntactic and semantic errors without manual, problem specific correction strategies. Given an incorrect student program, it generates candidate programs from a distribution of likely corrections, and checks each candidate for correctness against a test suite.