CAREER: Advancing Neural Testing and Debugging of Software

软件是每个人生活中不可或缺的一部分，从每个人口袋里的手机到已经行驶了数百万英里的自动驾驶汽车，再到实现智能家电的嵌入式软件。开发人员很容易犯错误并向软件引入错误，因此自动化软件验证技术对于确保交付可靠的软件至关重要。软件测试是发现和修复软件缺陷的活动，是软件开发人员的一项重要活动。随着人工智能（AI）的出现以及机器学习（ML）在理解和预测代码中的错误模式方面的潜在力量，软件测试和调试逐渐转向基于学习的技术，即软件的神经测试和调试。该研究项目将通过利用人工智能解决自动化软件测试和调试中的基本挑战，并将为语义健壮和可解释的神经测试和调试开发新的见解。结合理论构建、经验数据驱动研究和工具构建，本研究旨在(1)设计语义鲁棒的代码神经模型，并开发高质量数据集生成的系统方法；(2)开发用于功能和非功能测试的深度测试预言机的几种技术；(3)设计用于提取和重用神经模型知识的解释技术，以统一测试和调试。这些总体思想可以使软件测试和调试更聪明、更快，显著影响研究人员和实践者提高软件质量的方式。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。

Software is an integral part of every life, from cell phones in everyone's pocket to autonomous cars that have already driven millions of miles to embedded software enabling smart home appliances. Developers are prone to making mistakes and introducing bugs to the software, making automated software validation techniques essential to ensure delivering reliable software. Software testing is the activity of finding and fixing software bugs is an important activity for software developers. With the advent of Artificial Intelligence (AI) and the potential power of Machine Learning (ML) in understanding and predicting bug patterns in code, software testing and debugging are gradually moving towards learning-based techniques, i.e., neural testing and debugging of software. This research project will address fundamental challenges in automated software testing and debugging by leveraging AI, and will develop new insights for semantically robust and interpretable neural testing and debugging.Combining theory building, empirical data-driven research, and tool building, this research aims to (1) design semantically robust neural models of code and develop systematic approaches for high-quality dataset generation, (2) develop several techniques to construct deep test oracles for functional and non-functional testing, and (3) design interpretation techniques for extracting and reusing the knowledge of neural models to unify testing and debugging. These overarching ideas can make software testing and debugging smarter and faster, significantly impacting how researchers and practitioners improve software quality.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.