Machine learning (ML) is nowadays embedded in several computing devices, consumer electronics, and cyber-physical systems. Smart sensors are deployed everywhere, in applications such as wearables and perceptual computing devices, and intelligent algorithms power our connected world. These devices collect and aggregate volumes of data, and in doing so, they augment our society in multiple ways; from healthcare, to social networks, to consumer electronics, and many more. To process these immense volumes of data, ML is emerging as the de facto analysis tool that powers several aspects of our Big Data society. Applications spanning from infrastructure (smart cities, intelligent transportation systems, smart grids, and to name a few), to social networks and content delivery, to e-commerce and smart factories, and emerging concepts such as self-driving cars and autonomous robots, are powered by ML technologies. These emerging systems require real-time inference and decision support; such scenarios, therefore, may use customized hardware accelerators, are typically bound by limited resources, and are restricted to limited connectivity and bandwidth. Thus, near-sensor computation and near-sensor intelligence have started emerging as necessities to continue supporting the paradigm shift of our connected world. The need for real-time intelligent data analytics (especially in the era of Big Data) for decision support near the data acquisition points emphasizes the need for revolutionizing the way we design, build, test, and verify processors, accelerators, and systems that facilitate ML (and deep learning, in particular) implemented in resource-constrained environments for use at the edge and the fog. As such, traditional von Neumann architectures are no longer sufficient and suitable, primarily because of limitations in both performance and energy efficiency caused especially by large amounts of data movement. Furthermore, due to the connected nature of such systems, security and reliability are also critically important. Robustness, therefore, in the form of reliability and operational capability in the presence of faults, whether malicious or accidental, is a critical need for such systems. Moreover, the operating nature of these systems relies on input data that is characterized by the four “V’s”: velocity (speed of data generation), variability (variable forms and types), veracity (unreliable and unpredictable), and volume (i.e., large amounts of data). Thus, the robustness of such systems needs to consider this issue as well. Furthermore, robustness in terms of security, and in terms of reliability to hardware and software faults, in particular, besides their importance when it comes to safety-critical applications, is also a positive factor in building trustworthiness toward these disrupting technologies from our society. To achieve this envisioned robustness, we need to refocus on problems such as design, verification, architecture, scheduling and allocation policies, optimization, and many more, for determining the most efficient, secure, and reliable way of implementing these novel applications within a robust, resource-constrained system, which may or may not be connected. This special issue, therefore, addresses a key aspect of fog and edge-based ML algorithms; robustness (as defined above) under resource-constraint scenarios. The special issue presents emerging works in how we design robust systems, both in terms of reliability as well as fault tolerance and security, while operating with a limited number of resources, and possibly in the presence of harsh environments that may eliminate connectivity and pollute the input data.

中文翻译：

---

客座社论：用于机器学习的健壮的资源受限系统

机器学习（ML）如今，它已嵌入多种计算设备，消费类电子产品和网络物理系统中。智能传感器被部署在可穿戴设备和感知计算设备等应用中的任何地方，而智能算法则为我们的互联世界提供动力。这些设备收集和汇总大量数据，并以此方式以多种方式增强了我们的社会。从医疗保健，社交网络到消费类电子产品等等。为了处理这些庞大的数据量，机器学习正逐渐成为实际上分析工具，可为我们的大数据社会的各个方面提供支持。应用范围从基础架构（智能城市，智能交通系统，智能电网等）到社交网络和内容交付，电子商务和智能工厂，以及新兴概念（如自动驾驶汽车和自动机器人），由ML技术提供支持。这些新兴系统需要实时推理和决策支持。因此，这样的场景可能使用定制的硬件加速器，通常受有限的资源约束，并且受限于有限的连接性和带宽。因此，近传感器计算和近传感器智能已开始出现，以继续支持我们互联世界的范式转变。在数据采集点附近需要实时智能数据分析（尤其是在大数据时代）以提供决策支持，这强调了需要彻底改变我们设计，构建，测试和验证处理器，加速器和系统的方法，在资源受限的环境中实施的ML（尤其是深度学习）可在边缘和雾中使用。因此，传统的冯·诺依曼架构不再足够和合适，这主要是由于性能和能源效率方面的局限性，尤其是由于大量数据移动所致。此外，由于此类系统的连接特性，安全性和可靠性也至关重要。因此，在出现故障时以可靠性和操作能力的形式实现鲁棒性，无论是恶意的还是偶然的，对于此类系统都是至关重要的。此外，这些系统的操作性质取决于以四个“ V”为特征的输入数据：速度（数据生成的速度），可变性（可变形式和类型），准确性（不可靠且不可预测）和体积（即，大量数据）。因此，此类系统的鲁棒性也需要考虑这个问题。此外，在安全性方面，尤其是在硬件和软件故障的可靠性方面，鲁棒性，尤其是在对安全性至关重要的应用程序中的重要性时，也是建立对这些破坏性技术的信任度的积极因素。为了达到预期的鲁棒性，我们需要重新关注设计，验证，架构，调度和分配策略，优化等，以确定在健壮，资源受限的系统中实现这些新颖应用程序的最有效，安全和可靠的方式，该系统可能连接也可能不连接。因此，这个特殊问题解决了雾和基于边缘的ML算法的关键方面。资源受限情况下的鲁棒性（如上定义）。本期专刊介绍了新兴的作品，说明了我们在可靠性，容错性和安全性方面如何设计健壮的系统，同时使用的资源数量有限，并且可能存在恶劣的环境，这些环境可能会消除连接性并污染系统。输入数据。在健壮，资源受限的系统中实现这些新颖应用程序的可靠方式，该系统可能会连接也可能不会连接。因此，这个特刊解决了雾和基于边缘的ML算法的关键方面。资源受限情况下的鲁棒性（如上定义）。本期专刊介绍了新兴的作品，说明了我们在可靠性，容错性和安全性方面如何设计健壮的系统，同时使用的资源数量有限，并且可能存在恶劣的环境，这些环境可能会消除连接性并污染系统。输入数据。在健壮，资源受限的系统中实现这些新颖应用程序的可靠方式，该系统可能会连接也可能不会连接。因此，这个特殊问题解决了雾和基于边缘的ML算法的关键方面。资源受限情况下的鲁棒性（如上定义）。本期专刊介绍了新兴的作品，说明了我们在可靠性，容错性和安全性方面如何设计健壮的系统，同时使用的资源数量有限，并且可能存在恶劣的环境，这些环境可能会消除连接性并污染系统。输入数据。