“The Algorithm Distillation (AD) method can distill an algorithm into a neural network by performing behavioral cloning over a set of learning histories generated by source policies.”