The algorithm takes advantage of a probabilistic approach the researchers call “Bayesian Program Learning,” or BPL. Essentially, the computer generates its own additional examples, and then ...