Human-Neural Net Collaborative Programming

Current program generation models suffer from the challenges of poor generalization, narrow program coverage and being incompetent to dealing with complex program structures (e.g., branch, loop, recursion.), because these models only get single information input (input-output samples) and rely entirely on neural networks. Obviously, there is ambiguity to reverse program behaviors through input-output pairs alone. Thus, we propose a Human-Neural Net Collaborative Programming (HNCP) paradigm, that integrates the strengths of human's experience and perception with the advantages of neural network's automatic learning from data, in which programmers compose the overall program framework, but let the neural network learn to generate the local trivial dirty detail. The advantages are to seamlessly integrate neural network components as a construction of advanced programming language, and to promote the practical application of program generation. Experiments indicate that our approach is effective and outperforms the state-of-the-art in program generation.