ECCS-EPSRC - ShiRAS: Towards Safe and Reliable Autonomy in Sensor Driven Systems

现代传感器产生大量数据。数据驱动的算法，能够自我训练或“自我调整”，已经彻底改变了自治系统的领域。然而，从异构大规模数据的集成捕获信心仍然是这些算法的一个具有挑战性的任务。我们的工作将开发开创性的方法，在传感器驱动系统的不同层面上引入安全可靠的自主性。主要重点是机器学习方法与量化的不确定性或置信界限提供的解决方案。这项研究将通过正式的开发，分析和评估所提出的方法，从而产生安全可靠的机器智能，需要重要的理论知识。可扩展、有效和强大的算法将可用于计算智能中最关键的挑战之一。理解和评估现代机器学习模型的不确定性具有重要意义，特别是当这些模型的输出被输入到更高级别的决策过程中时。其中包括自动无人机和车辆，医疗领域的诊断和监控。本研究的案例研究和应用包括武器和违禁品检测领域的行业和政府合作伙伴（与美国国土安全部运输安全实验室合作），（美国联邦航空管理局）、智能交通系统（英国高速公路英格兰，Valerann有限公司英国和TransDAC，美国），以及智能城市和监控（QinetiQ有限公司英国，泰雷兹有限公司）。开源软件库，基于最先进的深度学习框架，例如，TensorFlow和PyTorch将用于快速传播已开发的核心计算技术。拟议的努力还包括将研究纳入本科生和研究生课程，制定外展一揽子计划，该奖项反映了NSF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进行评估，被认为值得支持的搜索.

Modern sensors generate massive amounts of data. Algorithms that are data-driven, able to train themselves or "self-tune", have revolutionized the area of autonomous systems. However, capturing confidence from the integration of heterogeneous large-scale data remains a challenging task for these algorithms. Our work will develop pioneering approaches that will introduce safe and reliable autonomy at different levels in sensor-driven systems. The main focus is on machine learning methods with quantified uncertainty or confidence bounds for the provided solutions. This research will entail significant theoretical knowledge through formal development, analysis and evaluation of the proposed approaches, resulting in safe and reliable machine intelligence. Scalable, effective and robust algorithms will be available for one of the most critical challenges in computational intelligence. Understanding and assessing the uncertainty of modern machine learning models has critical consequences, especially when the output of such models is fed into higher-level decision making processes. These include autonomous drones and vehicles, diagnosis in the medical domain and surveillance. Case studies and applications of this research include industry and government partners in the areas of detection of weapon and contraband (in collaboration with the Transportation Security Laboratory at the US Department of Homeland Security), rotorcraft safety (US Federal Aviation Administration), intelligent transportation systems (UK Highways England, Valerann Ltd UK, and TransDAC, US), and smart cities and surveillance (QinetiQ Ltd UK, Thales Ltd). Open-source software libraries, based on state-of-the-art deep learning frameworks, e.g., TensorFlow and PyTorch, will be built for the rapid dissemination of the developed core computing techniques. The proposed effort also includes integrating the research into the undergraduate and graduate curriculums, developing outreach packages which deliver hands-on experiences covering the fields of autonomous monitoring and the issues facing future intelligent transportation systems and the cities as a whole.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.