Reviews: Neural Guided Constraint Logic Programming for Program Synthesis

The paper presents an approach to neural guided search for program synthesis of LISP programs. A modified miniKanren constraint solver is used to synthesize a program from example IO pairs by "inverting" the EVAL function. The search for candidate programs in the solver is guided by a neural network (either a GNN or an RNN with other layers on top). The approach is compared to baselines and three symbolic methods. Pros: - The paper is very well written and definitely relevant to NIPS - The evaluation against symbolic methods is reasonable - While the idea of scoring evaluation expansions is not novel, applying it to the EVAL function is Cons: - Novelty is somewhat limited - Links to some recent related work are missing - No comparison to other neural methods In general, I think that paper is clear and sound enough.