引用EvgeniyZh · 2017-02
“That's interesting that the paper was on arxiv since November and people payed attention just now.”
指出 DeepCoder 论文 11 月就挂在 arXiv 上,直到现在才受关注
用神经网络预测程序性质,加速基于输入输出示例的程序搜索
针对由输入输出样例求解编程竞赛风格题目的问题,训练神经网络从样例预测生成该输出的程序具有哪些性质,再用预测结果增强编程语言领域的枚举搜索和基于 SMT 的求解器。实验显示比未增强的强基线和 RNN 方法快一个数量级,能解出难度相当于竞赛网站上最简单题目的问题。
评论者在讨论神经网络生成代码时把它当作神经程序合成的代表性例子,也有人提到它早在前一年 11 月就已挂上 arXiv。
“That's interesting that the paper was on arxiv since November and people payed attention just now.”
“Note that we haven't yet gotten a good neural-generated code modifications, even using large networks, GPU training and large computing time. Best examples I could find:”