Software Engineering for IoT Data-Driven Machine Learning Applications

许多组织越来越多地采用人工智能(AI)，特别是机器学习(ML)，具有为行业和社会带来巨大价值的潜力。然而，我们在由此产生的软件系统中遇到了故障和缺陷。主要原因是AI和ML产生的开发范式的转变。虽然ML任务通常与ML算法和技术的开发和增强有关，但软件工程(SE)主要关注软件需求规范、测试、部署和演化。ML应用程序与传统软件的不同之处在于，它们的逻辑不是显式编程的，而是通过不断学习数据来自动创建的。基于ML的系统开发过程涉及不同的活动，包括数据收集、培训和模型评估。这些任务主要由ML和领域专家执行，少量由软件工程师执行。然而，由于ML系统开发的不同方法，整个系统的开发需要针对已经建立的SE过程的新方法，或者可能是全新的方法。此外，我们目前正在见证涉及高速和大容量数据流的新一代软件应用，例如与物联网(IoT)相关的软件应用。物联网是将任何设备连接到互联网，实现设备、车辆和其他现实世界元素的数字化和基于服务的协调。因此，从远程医疗到智能交通、智能电网和工业4.0，物联网被认为是影响社会和行业的各种形式的重大范式转变也就不足为奇了。所有这些应用都需要高效地获取、处理和管理高速、海量的数据，然后才能用于ML模型的训练和推理。该研究计划旨在为物联网数据驱动的机器学习应用程序的建模、设计和部署带来实质性的进步。在基础研究层面，重点将放在为设计、发展和部署实用的基于ML的系统建立新颖、合理的SE方法上。研究结果将使软件工程师能够将经过ML培训的模型转换为行业强度的生产质量ML应用程序。这项研究将通过提供一种结构化的方法来开发处理物联网数据的ML应用程序，从而在自动化过程中使加拿大的各种行业(如智能工厂、智能建筑、先进制造)受益。拟议的研究计划还将提供巨大的HQP培训机会，并为加拿大行业配备使用物联网数据工程ML应用程序的专家，据信这一职业在未来五到十年内将继续保持旺盛需求。

Many organizations have increasingly adopted artificial intelligence (AI), particularly machine learning (ML), with the potential to deliver enormous value to industry and society. However, we experience failures and shortcomings in the resulting software systems. The main reason is the shift in the development paradigm generated by AI and ML. While ML tasks are typically related to the development and enhancement of ML algorithms and techniques, software engineering (SE) primarily focuses on the software requirements specification, testing, deployment, and evolution. ML applications differ from traditional software in that their logic is not explicitly programmed but instead automatically created by continuously learning from data. The ML-based systems' development process involves different activities, including data collection, training, and model evaluation. These tasks are mostly performed by ML and domain experts and, to a lesser extent, by software engineers. Nevertheless, due to the different approaches in the development of ML systems, the entire system's development requires new methods for already established SE processes or possibly wholly new approaches. In addition to the above, we are currently witnessing a new generation of software applications which involve high speed and high volume data streams, such as software applications related to the Internet of Things (IoT). IoT is about connecting any device to the Internet, enabling the digitization and service-based coordination of devices, vehicles, and other real-world elements. Therefore, it is no surprise that IoT has been recognized as a significant paradigm shift impacting both society and industry in numerous forms ranging from telemedicine to smart transportation, smart grids, and Industry 4.0. All these applications require efficient acquisition, processing, and management of high speed and high volume data before being used for training and reasoning by ML models. This research program aims to bring substantial advancements in modeling, designing, and deploying IoT data-driven machine learning applications. At the fundamental research level, the focus will be on establishing novel, and sound SE approaches for designing, evolving, and deploying practical ML-based systems. The research results will enable software engineers to transition ML-trained models to industry-strength production-quality ML applications. This research will benefit a diverse Canadian industry, such as smart factories, smart buildings, advanced manufacturing, in their automation processes by providing a structured approach to develop ML applications dealing with IoT data. The proposed research program will also stage a tremendous HQP training opportunity and equip Canadian industries with experts in engineering ML applications using IoT data, a profession believed to continue to be in high demand in the next five to ten years.