China
Africa
Explore chapters and articles related to this topic
Hardware Realization of Reinforcement Learning Algorithms for Edge Devices
Published in Sandeep Saini, Kusum Lata, G.R. Sinha, VLSI and Hardware Implementations Using Modern Machine Learning Methods, 2021
Shaik Mohammed Waseem, Subir Kumar Roy
Though there are several edge-based computing frameworks that have been developed by companies like Baidu (OpenEdge) [12], Microsoft (Azure IoT Edge) [13], and Amazon (AWS IoT Greengrass) [14], the work is continuously expanding to accommodate more flexible and compatible frameworks and make possible computing at the edge for several day-to-day applications and create a considerable impact on solving several societal issues and problems. There exist several deep learning libraries with support for edge computations. A few of the popular ones include PyTorch from Facebook [15], Apple’s CoreML [16], Qualcomm’s SNPE [17], Paddle-Mobile from Baidu [18], and MACE from XiaoMi [19]. Google has two versions as part of deep learning libraries, Tensorflow [20] is heavy weight and therefore unsuitable and generally not recommended for the edge. A lighter version called Tensorflow Lite [21] has been developed by Google for mobile/edge devices. Though these libraries have established themselves as potential choices, there exist several other open-source deep learning libraries that can serve equally well and therefore are available for consideration.
Edge-Based Blockchain Design for IoT Security
Published in Sudhir Kumar Sharma, Bharat Bhushan, Aditya Khamparia, Parma Nand Astya, Narayan C. Debnath, Blockchain Technology for Data Privacy Management, 2021
Pao Ann Hsiung, Wei-Shan Lee, Thi Thanh Dao, I. Chien, Yong-Hong Liu
In edge computing, data is processed near the data collection source, so there is no longer a need to transfer the data to the cloud or to a local data center for processing and analysis. This approach reduces the load on both network and cloud. Due to its ability to process data in real time, and its faster response time, edge computing has a high applicability in the IoT field, especially in the Industrial Internet of Things (IIoT). In addition to accelerating the digital transformation of industrial and manufacturing companies, edge computing technologies can also enable innovations including artificial intelligence and machine learning. However, edge computing also faces the problem of deployment, specifically, how to effectively deploy the subordinates at various nodes. In 2017, Rakesh Jain and Samir Tata [25] proposed a deployment method using RED-Node. In this work, a dynamically reconfigurable edge computing architecture is proposed for the IoT based on Docker containers that are automatically orchestrated using Kubernetes.
Edge Computing and Embedded Storage
Published in Nishu Gupta, Joel J. P. C. Rodrigues, Justin Dauwels, Augmented Intelligence Toward Smart Vehicular Applications, 2020
Ramkumar Jayaraman M. Baskar, B. Amutha
Edge devices, such as IoT sensors, computers, and smartphones, have security-enhanced devices with internet-based devices which are included in the edge-based infrastructure, as represented in Figure 2.2. With the help of edge gateway, data can be processed from the IoT devices and then sends only the relevant information through the cloud location, which reduces the bandwidth. After data gets processed, data can also be sent back to IoT devices for real-time applications. Thus, the edge computing process takes place with specific components including IoT devices, local edge processing, edge gateway, cloud-based location, and data center. Based on edge computing components, edge technology architecture can be deployed in various companies to save cost. As many companies are concentrating on reducing bandwidth cost and increasing performance, they are adopting cloud platforms in real-time applications [15].
Evolutionary PSO-based emergency monitoring geospatial edge service chain in the emergency communication network
Published in International Journal of Digital Earth, 2023
Sheng He, Xicheng Tan, Yanfei Zhong, Meng Huang, Zhiyuan Mei, You Wan, Huaming Wang
Meanwhile, the development of edge computing technology provides a new way for geographic information services to support DER in emergency communication networks. Different from the centralized and unified processing model of cloud computing, edge computing is a new model that performs computing tasks decentralized at the edge nodes of the network (Shi et al. 2016). Without uploading all the data to the cloud computing center, edge computing has higher data processing efficiency and is more suitable for time-sensitive application scenarios. Moreover, edge computing well evades and solves the problems of over-reliance on communication capabilities and data security in the cloud computing model (Liao and Wu 2020; Wang et al. 2020). In emergencies, edge nodes in LAN can still operate in an orderly manner, and almost no communication with the cloud is required to complete the computing task, which is of great practical importance. However, how to build efficient and reliable geospatial edge service (GES) chains for emergency monitoring tasks in the edge computing environment based on emergency communication networks is still a big challenge because of the spatiotemporal dynamicity of the edge computing environment and the emergency communication networks.
A cyber physical production system framework for online monitoring, visualization and control by using cloud, fog, and edge computing technologies
Published in International Journal of Computer Integrated Manufacturing, 2023
Rishi Kumar, Kuldip Singh Sangwan, Christoph Herrmann, Sachin Thakur
The necessity for implementing edge computing comes from the fact that due to digitalization, an increasing amount of data is generated at the edge of the network. If these data get processed at the edge of the network, then it would be more efficient, particularly for applications such as autonomous driving, healthcare monitoring, and process control systems where service latency management is significant for providing high quality-of-service and experience to terminal users (Shi et al. 2016; Cao et al. 2021; Blesson et al. 2016). Therefore, offloading some computing tasks at the edge of network is more efficient in terms of handling enormous raw data, energy savings, and reduced response time. Edge computing, where data processing is performed at the closer proximity to data sources, ensures shorter response time, cost saving with shorter bandwidth usage, data safety and privacy, better battery life, and energy saving (Shi et al. 2016). These three innovative and complementary computing technologies (cloud, fog, and edge) can be instrumental in utilizing the full potential of CPPS implementation by complementing to meet the specific requirements like latency, bandwidth, energy efficiency, security, etc., for online monitoring, visualization, and control.
Tensor decomposition to compress convolutional layers in deep learning
Published in IISE Transactions, 2022
Yinan Wang, Weihong “Grace” Guo, Xiaowei Yue
Given the properties of our proposed CPAC-Conv layer, we can summarize its contributions to industrial applications into three aspects. First, edge computing is becoming increasingly used in manufacturing systems, in which the edge devices carry out a substantial amount of tasks but usually have limited computational resources. Compressing the model complexity will potentially reduce the computational requirements for Deep Neural Networks implementation in edge devices. Second, our proposed CPAC-Conv layer is an alternative to replace the original Conv layer. It is not limited to process image data. It can also be used to compress the Deep Neural Networks with Conv layers for processing videos and point clouds. Our proposed CPAC-Conv layer can be used as one alternative building block for designing novel deep learning methods. Lastly, the model with fewer parameters is more promising according to the Occam’s Razor principle. It can have better interpretability and help us to understand the relationship between the decomposed kernels and extracted features.