Neural Guided Constraint Logic Programming for Program Synthesis

Abstract

Synthesizing programs using example input/outputs is a classic problem in artificial intelligence. We present a method for solving Programming By Example (PBE) problems by using a neural model to guide the search of a constraint logic programming system called miniKanren. Crucially, the neural model uses miniKanren's internal representation as input; miniKanren represents a PBE problem as recursive constraints imposed by the provided examples. We explore Recurrent Neural Network and Graph Neural Network models. We contribute a modified miniKanren, drivable by an external agent, available at https://github.com/xuexue/neuralkanren. We show that our neural-guided approach using constraints can synthesize programs faster in many cases, and importantly, can generalize to larger problems. 1 The name "Kanren" comes from the Japanese word for "relation".

miniKanren as a Tool for Symbolic Computation in Python

“…ese conde idioms comprise a large portion of the miniKanren work implied here, and their size could grow very quickly over time. is leads to performance questions that are possibly answered by work on guided search (Swords and Friedman;Zhang et al 2018) and discerning conde branch selection (Boskin et al 2018).…”

“…Finally, we would like to point out the potential for an exciting "feedback loop": as statistical modeling improves the processing of miniKanren (Zhang et al 2018), miniKanren can also improve the process of statistical modeling.…”