LearningCompositionalRulesviaNeuralProgram Synthesis

Many aspects of human reasoning, including language, require learning rules from very little data. Humans can do this, often learning systematic rules from veryfewexamples,andcombining theserulestoformcompositional rule-based systems.