STUNT - Improving Software Testing Using Novelty

软件测试是识别错误和确保软件质量的基本活动。因为测试通常是乏味且容易出错的，所以自动化的测试生成旨在支持开发人员和测试人员找到最有效的测试用例。已经提出了许多不同的测试生成技术，其中一些最流行的是由源代码指导的技术（例如，白盒技术）。然而，经典的白盒技术（如代码覆盖率）是否真的与故障查找效率相关，在研究中是一个备受争议的话题，并且在测试生成过程中，源代码度量所提供的指导是粗糙和不充分的，从而抑制了有效的测试。这个项目的目的是提高自动化测试生成的有效性，通过使用新颖性的概念来评估程序行为，而不是源代码，以更好地指导测试生成，并产生更有效的测试套件。该项目将解决量化新颖性的问题，并将使用机器学习技术根据程序行为来推理新颖性，而不仅仅是根据传统的白盒测量。新颖性的概念将被集成到有效的测试生成算法中，用新颖性的概念增强最先进的白盒技术，从而产生更有效的测试套件。

Software testing is an essential activity for identifying faults and ensuring software quality. Because testing generally is tedious and error-prone, automated test generation aims to support developers and testers in finding the most effective test cases. Many different techniques for test generation have been proposed, and some of the most popular ones are those guided by the source code (i.e., white-box techniques). However, whether classic white- box techniques such as code coverage really correlate with fault finding effectiveness is a much debated topic in research, and the guidance provided by source code metrics during test generation is known to be coarse and insufficient, thus inhibiting effective testing. The aim of this project is to improve the effectiveness of automated test generation by using the concept of novelty evaluated on program behaviour rather than source code to better guide test generation, and to produce more effective test suites. The project will address the problem of quantifying novelty, and will use machine learning techniques to reason about novelty in terms of program behaviour, rather than just in terms of traditional white-box measurements. The concept of novelty will be integrated into effective test generation algorithms, enhancing state-of-the-art white-box techniques with the concept of novelty, thus resulting in more effective test suites.