We propose Bee Search, a best-first bottom-up search algorithm for program synthesis that uses a learned cost function to guide the search.

We introduce POLIS, a local search method for program optimization that iteratively improves programs by applying local transformations.