Automatic bug-repair system fixes 10 times as many errors as its predecessors

MIT researchers have developed a machine-learning system that can comb through repairs to open-source computer programs and learn their general properties, in order to produce new repairs for a different set of programs.