Microsoft Research's DeepCoder and IBM Research's efforts to scan peer-reviewed papers for code, framework, and library details are examples of attempts to automate code generation. However, these efforts face limitations such as the inability to generate large quantities of code at a time, and the lack of standard frameworks or libraries used widely enough by the developer community to have auto-code generating power. Despite these challenges, researchers are working on solutions, such as creating a "grammar" for an abstract representation that can be platform agnostic, and predicting layers for neural networks to increase productivity.
Key takeaways:
- Auto-code generation is not a new concept, but it has been getting fresh attention due to better capabilities and ease of use in neural network frameworks.
- Google, Microsoft and IBM have announced new ways of boosting developer productivity with deep learning frameworks that fill themselves in—at least in part.
- Microsoft Research's DeepCoder and IBM Research's DLPaper2Code are examples of efforts to automate code generation, but they face limitations due to data input and the need for standardization.
- While there is progress towards greater auto-code generation, the reality of fully automated programming is still quite a long way off.