SBIR Phase I: The Development of an Artificial Analysis (AI) Static Code Analysis Platform to Increase Software Developer Productivity

这个小企业创新研究（SBIR）第一阶段项目的更广泛/商业影响是在广泛的行业中提高软件开发的速度和准确性，并使软件开发人员更富有成效。该技术将通过交付更高质量和无缺陷的代码来减少花费在审查代码上的时间，并将进一步确保更安全的软件不容易受到外部攻击。SBIR一期项目开发了一种基于云的基于人工智能（AI）的静态代码分析工具，可以在软件开发过程的早期发现复杂和严重的问题。与现有的静态分析工具不同，本项目开发的工具将自动从错误修复中学习，解释发现的错误，并就如何修复它们提出建议。结果将有助于金融、医疗保健和国防行业的组织和开发人员，在这些行业中，代码重用对于安全性和遵从性非常重要。总的来说，这个项目很好地适应了组织将更多的人工智能集成到他们的操作和不断增长的软件开发工具市场的趋势。SBIR第一阶段项目结合了机器学习和自然语言处理的最新进展，开发了一种新的智能方法来发现和解释软件错误。该项目关注于开发一个软件架构，该架构能够分析完整的模型层次结构，建立一种技术来有效和定量地评估为标记错误生成的解释的有效性，并将不同的组件集成到单个分析框架中。该项目将包括三个模型，这些模型将作为总体系统架构的一部分进行开发和集成：(1)利用图形注意网络的代码故障检测模型，(2)生成转换器以构建解释和建议，以及(3)生成图对图转换器以生成代码架构的突变以解决标记的错误。该项目将利用基于变压器和基于图的神经网络的最新进展，从而推动当前高效代码审查过程的研究。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase I project is to increase the speed and accuracy of software development in a wide range of industries and to make software developers more productive. The technology will decrease the time spent reviewing code by shipping higher quality and defect free code and will further ensure more secure software that is less prone to outside attacks. This SBIR Phase I project develops a cloud-based artificial intelligence (AI)-based static code analysis tool which can find complex and severe problems early in the process of software development. Unlike existing static analysis tools, the tool developed in this project will learn automatically from bug fixes, explain the errors found, and make recommendations on how to fix them. Results will help organizations and developers in the finance, healthcare, and defense industries where code reuse is important for security and compliance reasons. Overall, this project fits well with an increasing trend of organizations integrating more AI into their operations and a growing market for software development tools.This SBIR Phase I project combines the latest advancements in machine learning and natural language processing to develop a new, intelligent way to find and explain software errors. The project focuses on developing a software architecture that enables the analysis of a complete model hierarchy, establishing a technique to effectively and quantitatively evaluate the validity of explanations generated for flagged bugs, and integrating the disparate components into a single analysis framework. The project will consist of three models which will be developed and integrated as part of the overarching system architecture: (1) a code fault detection model utilizing a graph attention network, (2) a generative transformer to build explanations and suggestions, and (3) a graph-to-graph transformer to generate mutations to the code architecture to resolve the flagged bugs. The project will leverage recent advancements in transformer-based and graph-based neural networks and therefore propel the current state of research for efficient code review processes forward.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.