Testing Deep Neural Network Programs

测试深度神经网络程序

• 批准号: 542649-2019
• 负责人: Hemmati, Hadi
• 金额: $ 1.82万
• 依托单位: University of Calgary
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2019
• 资助国家: 加拿大
• 起止时间: 2019-01-01 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=681211
• 关键词: Testing; Deep; Neural; Network; Programs

In recent years, artificial intelligence and in particular Machine Learning (ML) have been the core of research and development in most companies around the world. Techniques such as deep learning have been employed in many contexts and provided astonishing results. However, as we employ these ML techniques on a variety of applications and contexts, our everyday life becomes more dependent on their performance. For example, in the context of safety critical or mission critical applications a reliable, robust, safe, and secure ML is a must. Thus, in this project, we propose studying techniques for testing ML programs (particularly in the context of object detection in videos using deep learning) to understand how good the existing techniques are in terms of detecting real defects and how we can improve those techniques with respect to their shortcomings in our case study with OASIS Inc company.

近年来，人工智能，特别是机器学习（ML）一直是全球大多数公司研究和开发的核心。深度学习等技术已经在许多环境中使用，并提供了惊人的结果。然而，随着我们在各种应用程序和环境中使用这些ML技术，我们的日常生活变得更加依赖于它们的性能。例如，在安全关键型或使命关键型应用的背景下，可靠、健壮、安全和安全的ML是必须的。因此，在这个项目中，我们建议研究用于测试ML程序的技术（特别是在使用深度学习进行视频中的对象检测的背景下），以了解现有技术在检测真实的缺陷方面的效果，以及我们如何改进这些技术，以弥补OASIS Inc公司案例研究中的缺点。