Conference Paper

Edge and Fog Computing Enabled AI for IoT-An Overview

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... In recent years, researchers have increasingly turned to AI to help them analyse large amounts of data for the aforementioned uses. AI's Machine Learning (ML) and Deep Learning (DL) subfields give useful data insights and decision help [72,73]. Following that, we are discussing some of the AI-enabled fog computing technologies that make these applications possible. ...
... For the IoT, 5G signifies more than just a new era of wireless innovation. More than trillions of sensors, gadgets, and machines are powered by AI and run autonomously from the data centre to the edge of the network [72,73]. In terms of speeding up data analysis and decision making, fog computing and edge computing are the two best technologies. ...
... • Fault-tolerance and availability: One of the primary reasons for the development of fog computing was to improve dependability. When it comes to fog computing, difficulties like sensor failure, a lack of access network coverage in a particular region or the entire network, service platform failure, and a broken user interface system connection are all part of the equation [72]. Another challenge in the fog environment is to increase the availability of apps. ...
Article
Full-text available
Autonomic computing investigates how systems can achieve (user) specified “control” outcomes on their own, without the intervention of a human operator. Autonomic computing fundamentals have been substantially influenced by those of control theory for closed and open-loop systems. In practice, complex systems may exhibit a number of concurrent and inter-dependent control loops. Despite research into autonomic models for managing computer resources, ranging from individual resources (e.g., web servers) to a resource ensemble (e.g., multiple resources within a data center), research into integrating Artificial Intelligence (AI) and Machine Learning (ML) to improve resource autonomy and performance at scale continues to be a fundamental challenge. The integration of AI/ML to achieve such autonomic and self-management of systems can be achieved at different levels of granularity, from full to human-in-the-loop automation. In this article, leading academics, researchers, practitioners, engineers, and scientists in the fields of cloud computing, AI/ML, and quantum computing join to discuss current research and potential future directions for these fields. Further, we discuss challenges and opportunities for leveraging AI and ML in next generation computing for emerging computing paradigms, including cloud, fog, edge, serverless and quantum computing environments.
... In recent years, researchers have increasingly turned to AI to help them analyse large amounts of data for the aforementioned uses. AI's Machine Learning (ML) and Deep Learning (DL) subfields give useful data insights and decision help [72,73]. Following that, we are discussing some of the AI-enabled fog computing technologies that make these applications possible. ...
... For the IoT, 5G signifies more than just a new era of wireless innovation. More than trillions of sensors, gadgets, and machines are powered by AI and run autonomously from the data centre to the edge of the network [72,73]. In terms of speeding up data analysis and decision making, fog computing and edge computing are the two best technologies. ...
... • Fault-tolerance and availability: One of the primary reasons for the development of fog computing was to improve dependability. When it comes to fog computing, difficulties like sensor failure, a lack of access network coverage in a particular region or the entire network, service platform failure, and a broken user interface system connection are all part of the equation [72]. Another challenge in the fog environment is to increase the availability of apps. ...
Preprint
Full-text available
Autonomic computing investigates how systems can achieve (user) specified control outcomes on their own, without the intervention of a human operator. Autonomic computing fundamentals have been substantially influenced by those of control theory for closed and open-loop systems. In practice, complex systems may exhibit a number of concurrent and inter-dependent control loops. Despite research into autonomic models for managing computer resources, ranging from individual resources (e.g., web servers) to a resource ensemble (e.g., multiple resources within a data center), research into integrating Artificial Intelligence (AI) and Machine Learning (ML) to improve resource autonomy and performance at scale continues to be a fundamental challenge. The integration of AI/ML to achieve such autonomic and self-management of systems can be achieved at different levels of granularity, from full to human-in-the-loop automation. In this article, leading academics, researchers, practitioners, engineers, and scientists in the fields of cloud computing, AI/ML, and quantum computing join to discuss current research and potential future directions for these fields. Further, we discuss challenges and opportunities for leveraging AI and ML in next generation computing for emerging computing paradigms, including cloud, fog, edge, serverless and quantum computing environments.
... Although sometimes the edge and fog computing terms can be used interchangeably by authors, due to the lack of a standardized accepted definition, we can describe them as follows [3], [4]: ...
... Cloud computing generally demands high latency, connectivity dependency, and network bandwidth, which can be considered a waste of resources (energy and bandwidth, for instance) [5]. In this sense, we can list some edge and fog computing improvements compared to cloud computing: decreased response times (80 to 200 ms), overall service latency (50%), and energy consumption (30 to 40%) [4]. A comparison among edge and cloud, measuring the communication and processing latency, can show that the edge can be better thanks to the faster communication latency, even considering that the cloud servers can have faster processing with more robust servers [6]. ...
... The authors present a systematic classification, summarizing experiences from the surveyed works and suggesting directions for future research. Zou et al. [4] presented the main characteristics of edge and fog computing, summarizing the challenges regarding the enablement of AI for edge/fog-based IoT scenarios. One of the main challenges for edge devices running complex AI algorithms is decreasing resources consumption, improving energy efficiency. ...
Article
Full-text available
With the increasing adoption of the edge computing paradigm, including multi-access edge computing (MEC) in telecommunication scenarios, many works have explored the benefits of adopting it. Since MEC, in general, presents a reduction in latency and energy consumption compared to cloud computing, it has been applied to deploy artificial intelligence services. This kind of service can have distinct requirements, which involve different computational resource capabilities as well different data formats or communication protocols to collect data. In this sense, we propose the VEF Edge Framework, which aims at helping the development and deployment of artificial intelligence services for MEC scenarios considering requirements as low-latency and CPU/memory consumption. We explain the VEF architecture and present experimental results obtained with a base case's implementation: an object detection inference service deployed with VEF. The experiments measured CPU and memory usage for the VEF's main components and the processing time for two procedures (inference and video stream handling).
... Joint AI and F-RAN enhances network speed, reliability, and user satisfaction through dynamic and promising capabilities. Besides, F-RANs provide maximum spectral and energy efficiency, low latency, and high reliability to various distinct applications (e.g., mobile vehicular connectivity and smart industrial automation [5,6]). AI-driven F-RANs empower effective, fair, progressive, quick, and reliable functionality monitoring with promising features in smart fog-computing-based automation systems. ...
... Researchers in [4,5] examine the signal transmission and network traffic handling mechanisms accordingly for vehicular networks. Fog-and edge-driven IoT architecture and real-time smart city data traffic monitoring in a smart vehicular platform are discussed in [6,7], respectively. In [8] a latency optimization approach is proposed for software defined a cloud/edge vehicular platform. ...
Article
Artificial intelligence (AI)-driven fog computing (FC) and its emerging role in vehicular networks is playing a remarkable role in revolutionizing daily human lives. Fog radio access networks are accommodating billions of Internet of Things devices for real-time interactive applications at high reliability. One of the critical challenges in today's vehicular networks is the lack of standard wireless channel models with better quality of service (QoS) for passengers while enjoying pleasurable travel (i.e., highly visualized videos, images, news, phone calls to friends/relatives). To remedy these issues, this article contributes significantly in four ways. First, we develop a novel AI-based reliable and interference-free mobility management algorithm (RIMMA) for fog computing intra-vehicular networks, because traffic monitoring and driver's safety management are important and basic foundations. The proposed RIMMA in association with FC significantly improves computation, communication, cooperation, and storage space. Furthermore, its self-adaptive, reliable, intelligent, and mobility-aware nature, and sporadic contents are monitored effectively in highly mobile vehicles. Second, we propose a reliable and delay-tolerant wireless channel model with better QoS for passengers. Third, we propose a novel reliable and efficient multi-layer fog driven inter-vehicular framework. Fourth, we optimize QoS in terms of mobility, reliability, and packet loss ratio. Also, the proposed RIMMA is compared to an existing competitive conventional method (i.e., baseline). Experimental results reveal that the proposed RIMMA outperforms the traditional technique for intercity vehicular networks.
... On the other hand, portability is also relevant, and therefore, power will often come from an external battery or an energy harvesting subsystem, which imposes several challenges in the design of AI-enabled IoT devices. For example, in [97], a study regarding low-power ML architectures has been put forward and results have shown that sub-mW power consumption can potentially be deployed in "always-ON" AI-enabled IoT nodes. ...
... Conventional IoT devices are ubiquitous and low-cost, but natively resource-constrained, which limits their usage in ML tasks; however, data generated at the edge are increasingly being used to support applications that run ML models. Until now, edge ML has been predominantly focused on mobile inference, but recently several embedded ML solutions have been developed to operate in ultra-low-power devices, typically characterized by its hard resource constraints [97]. Recently, a new field of ML, known as Tiny ML, was put forward to enable inference at the edge endpoints. ...
Article
Full-text available
Internet of Things (IoT) can help to pave the way to the circular economy and to a more sustainable world by enabling the digitalization of many operations and processes, such as water distribution, preventive maintenance, or smart manufacturing. Paradoxically, IoT technologies and paradigms such as edge computing, although they have a huge potential for the digital transition towards sustainability, they are not yet contributing to the sustainable development of the IoT sector itself. In fact, such a sector has a significant carbon footprint due to the use of scarce raw materials and its energy consumption in manufacturing, operating, and recycling processes. To tackle these issues, the Green IoT (G-IoT) paradigm has emerged as a research area to reduce such carbon footprint; however, its sustainable vision collides directly with the advent of Edge Artificial Intelligence (Edge AI), which imposes the consumption of additional energy. This article deals with this problem by exploring the different aspects that impact the design and development of Edge-AI G-IoT systems. Moreover, it presents a practical Industry 5.0 use case that illustrates the different concepts analyzed throughout the article. Specifically, the proposed scenario consists in an Industry 5.0 smart workshop that looks for improving operator safety and operation tracking. Such an application case makes use of a mist computing architecture composed of AI-enabled IoT nodes. After describing the application case, it is evaluated its energy consumption and it is analyzed the impact on the carbon footprint that it may have on different countries. Overall, this article provides guidelines that will help future developers to face the challenges that will arise when creating the next generation of Edge-AI G-IoT systems.
... Many strategies exist for fog container deployment scheduling, ranging from simple but effective resource requests and grants [27], to using deep learning for allocation and real-time adjustments [28]. Initial research into fog computing and service scheduling dates from before the concept of the fog, for example Oppenheimer et al. [64], who studied migrating services in federated networks over large physical areas. This work takes into account available resources, network conditions, and the cost of migrating services between locations in terms of resources and latency. ...
... In a more general study, Zou et al. [64] list various hardware technologies that enable or accelerate specific types of AI in the edge. Most of these are designed for CNNs or deep learning in general, but some are aimed at Support Vector Machines (SVM). ...
... Cyber security management has been argued as one of the pillars in future manufacturing systems. To decrease latency while improving energy, efficiency, and security, processing data near or at the source of data has been highly demanded along with the cloud AI applications [77]. From the cloud services, information is exchanged over an enterprise service bus (EBS). ...
... Hence, dynamic process changes with shape recognition capabilities within the work cell are proposed to be monitored and detected locally by the feed of inspection cameras, FLIR cameras, and drone cameras. The processing power is preferably distributed near the data source at the edge devices [77] to further reduce the latency. An event-understanding system combined with ontological representation of manufacturing processes can be hereby described by the segmentation information and regional relationships among the binary masks, shown in Figure 15. ...
Article
Full-text available
A manufacturing paradigm shift from conventional control pyramids to decentralized, service-oriented, and cyber-physical systems (CPSs) is taking place in today’s 4th industrial revolution. Generally accepted roles and implementation recipes of cyber systems are expected to be standardized in the future of manufacturing industry. The authors intend to develop a novel CPS-enabled control architecture that accommodates: (1) intelligent information systems involving domain knowledge, empirical model, and simulation; (2) fast and secured industrial communication networks; (3) cognitive automation by rapid signal analytics and machine learning (ML) based feature extraction; (4) interoperability between machine and human. Semantic integration of process indicators is fundamental to the success of such implementation. This work proposes an automated semantic integration of data-intensive process signals that is deployable to industrial signal-based control loops. The proposed system rapidly infers manufacturing events from image-based data feeds, and hence triggers process control signals. Two image inference approaches are implemented: cloud-based ML model query and edge-end object shape detection. Depending on use cases and task requirements, these two approaches can be designated with different event detection tasks to provide a comprehensive system self-awareness. Coupled with conventional industrial sensor signals, machine vision system can rapidly understand manufacturing scenes, and feed extracted semantic information to a manufacturing ontology developed by either expert or ML-enabled cyber systems. Moreover, extracted signals are interpreted by Programmable Logical Controllers (PLCs) and field devices for cognitive automation towards fully autonomous industrial systems.
... In a variety of smart cities, AI has been widely deployed, yielding numbers of revolutionary applications and services that are primarily driven by techniques for data offloading for urban IoT [18,19]. The growing urbanization coupled with the high demands of daily commute for working professionals has increased the popularity of public transport systems [20,21]. ...
... Next, z 1 ′ ⋯ z t ′ is used as input to obtain k 1 ⋯ k t , and k 1 ⋯ k t is treated in the same way as z 1 ⋯ z t to obtain m 1 ⋯ m t . The above calculation process is shown in Equations (16), (17), (18), (19), and (20), where x 1 ⋯ x t is the input, z 1 ′ ⋯ z t ′ and k 1 ′ ⋯ k t ′ are z 1 ⋯ z t and k 1 ⋯ k t after the ReLu activation function, and m 1 ⋯ m t is the output. ...
Article
Full-text available
Internet of Things will play a vital role in the public transport systems to achieve the concepts of smart cities, urban brains, etc., by mining continuously generated data from sensors deployed in public transportation. In this sense, smart cities applied artificial intelligence techniques to offload data for social governance. Bicycle sharing is the last mile of urban transport. The number of the bike in the sharing stations, to be rented in future periods, is predicted to get the vehicles ready for deployment. It is an important tool for the implementation of smart cities using artificial intelligence technologies. We propose a DBSCAN-TCN model for predicting the number of rentals at shared bicycle stations. The proposed model first clusters all shared bicycle stations using the DBSCAN clustering algorithm. Based on the results of the clustering, the data on the number of shared bicycle rentals are fed into a TCN neural network. The TCN neural network structure is optimized. The effects of convolution kernel size and Dropout rate on the model performance are discussed. Finally, the proposed DBSCAN-TCN model is compared with the LSTM model, Kalman filtering model, and autoregressive moving average model. Through experimental validation, the proposed DBSCAN-TCN model outperforms the traditional three models in terms of two metrics, root mean squared logarithmic error, and error rate, in terms of prediction performance.
... Moreover, the rapid growth of CPSS data and resource constrained environment pose new problems to cross-modal retrieval service in real life applications [17], [18]. Nowadays, Cloud-Fog-Edge computing architecture is an effective way to address the above problems, which distributes the concentrated processes in a cloud via utilizing hierarchical computing resources [15], [16]. Data processing such as executing cross-modal retrieval near the source of data or at the edge is highly demanded. ...
Article
Full-text available
Data generated and collected from Cyber-Physical-Social Systems (CPSS) that usually in the forms of image, audio, video, and text, are complex and heterogeneous. How to deal with the crossmodal retrieval problem for heterogeneous CPSS data has has drawn considerable interests recently. The hashing based methods has been widely studied in building bilateral semantic associations of binary codes for cross-model retrieval. However, most existing methods discard binary constraints and learn linear projections as hashing functions. Moreover, none of them consider the cross-modal retrieval application in the scenario of Cloud-Fog-Edge computing. Therefore, how to learn more compact and discriminative binary codes with discrete constraints and nonlinear hashing functions for CPSS data in the Cloud-Fog- Edge architecture is still an open problem. In this paper, we propose a nonlinear discrete cross-modal hashing (NDCMH) method based on concise binary classification for CPSS data which fully investigates the nonlinear relationship embedding, discrete optimization as well as the hashing functions learning. Different from previous methods, our work presents a concise but promising cross-modal hashing method that builds a direct connection between original CPSS data and binary codes, which can alleviate the impact of large quantization loss. Furthermore, we execute the cross-modal retrieval service at cloud and fog. Specifically, hashing functions are deployed at the fog plane to reduce the amount data transfer and storage need on the cloud. Extensive experiments carried out on typical CPSS datasets demonstrate that the proposed NDCMH significantly outperforms other state-of-the-art methods.
... Active research areas in TIERS include multi-robot coordination [1], [2], [3], [4], [5], swarm design [6], [7], [8], [9], UWB-based localization [10], [11], [12], [13], [14], [15], localization and navigation in unstructured environments [16], [17], [18], lightweight AI at the edge [19], [20], [21], [22], [23], distributed ledger technologies at the edge [24], [25], [26], [27], [28], [29], edge architectures [30], [31], [32], [33], [34], [35], offloading for mobile robots [36], [37], [38], [39], [40], [41], [42], LPWAN networks [43], [44], [45], [46], sensor fusion algorithms [47], [48], [49], and reinforcement and federated learning for multi-robot systems [50], [51], [52], [53]. ...
... • Artificial Intelligence: Research related to AI applications are leading to increase in demand for fog computing [194]. Artificial intelligent system can integrate with the fog computing framework to improve the overall performance of the AI systems. ...
Article
Full-text available
There has been rapid development in the number of Internet of Things (IoT) connected nodes and devices in our daily life in recent times. With this increase in the number of devices, fog computing has become a well-established paradigm to optimize various key Quality of Service (QoS) requirements such as latency, bandwidth limitation, response time, scalability, privacy and security. In this paper, we present a systematic literature review of fog computing. This review article aims to classify recently published studies and investigate the current status in the area of fog computing. In this work, we have discussed the important characteristics of fog computing frameworks and identified various issues related to its architectural design, QoS metrics, implementation details, applications and communication modes. We have proposed taxonomy for fog computing frameworks based on the existing literature and compared the different research work based on taxonomy. Finally, various open research challenges and promising future directions are highlighted for further research in the area of fog computing.
... Applications and sensors of electronic devices are used to produce data, usually a massive amount of it. As a result, many companies must take responsibility for routinely maintaining vast volumes of data [2]. Companies currently need a dynamic information management architecture because of the transition to cloud computing and the benefits of this shift, such as scalability, availability, and pay-as-you-use features [3]. ...
Article
Full-text available
Internet of Things (IoT) systems usually produce massive amounts of data, while the number of devices connected to the internet might reach billions by now. Sending all this data over the internet will overhead the cloud and consume bandwidth. Fog Computing's (FC) promising technology can solve the issue of computing and networking bottlenecks in large-scale IoT applications. This technology complements cloud computing by providing processing power and storage to the edge of the network. However, it still suffers from performance and security issues. Thus, machine learning (ML) attracts attention for enabling FC to settle its issues. Lately, utilizing ML has been a growing trend to improve FC applications, like resource management, security, lessen latency, and power usage. Also, intelligent FC was studied to address industry 4.0, bioinformatics, blockchain, and vehicular communication system issues. Due to the ML vital role in the FC paradigm, this work will shed light on recent studies that utilized ML in an FC environment. Background knowledge about ML and FC was also presented. This paper categorized the surveyed studies into three groups according to the aim of ML implementation. These studies were thoroughly reviewed and compared using sum-up tables. The results showed that not all studies used the same performance metric except those worked on security issues. In conclusion, the proposed ML models' simulations are not sufficient due to the heterogeneous nature of the FC paradigm.
... IA systems require lots of sensory data to be gathered from IoT devices; the major requirements for IoT deployments are privacy, low-latency, and stable connectivity with low bandwidth consumption [1]. These can be met by utilizing local processing or computing at the network edge, which can decrease the service latency by 50% and also a decrease of the data transmitted towards the network core [6]. ...
Conference Paper
Full-text available
The IoT presents a possibility to create an innovative collabo-ration setup between human and robots within the Industry 4.0, leveraging concepts like Intelligence Augmentation (IA). This project aims to develop a safe human-robot collaboration (HRC) system. The system will feature a robotic arm and advanced sensors, inter-networked in an edge computing framework. It will also make use of computer vision techniques, enabling a seamless and safe operation. Finally, the system operation will be verified on a set of prespecified manufacturing tasks, demonstrating adequate safety measures for effective collaboration.
... -Fog computing: Fog computing is considered as a similar concept as the edge computing that pushes the computation and resources closer to the end devices/users. However, the fog computing can also be considered as a bigger and richer umbrella of resources/services and an edge can be smaller subset/unit of the fog with limited resources [67]. Fog computing have become vital in a number of smart environment based applications and presents various key features, such as low-latency, orchestration functionalities, faster data processing, and decision making [68]. ...
... Some possible approaches to enabling embedded AI on edge devices may consist of either developing dedicated hardware [14,15] or of leveraging low-, mid-and high-end embedded computing boards to execute ad hoc frameworks that implement a classification or a prediction algorithm targeting a specific application [16][17][18][19]. Other solutions rely either on architectures where the edge/fog layer is in charge to run ML and sensor fusion algorithms on the data collected from the edge devices [20][21][22][23][24], or on cloud services that automatically compile pre-trained ML inference models into representations which are more suitable for running in resource-constrained edge devices [25]. ...
Article
Full-text available
Low-cost, high-performance embedded devices are proliferating and a plethora of new platforms are available on the market. Some of them either have embedded GPUs or the possibility to be connected to external Machine Learning (ML) algorithm hardware accelerators. These enhanced hardware features enable new applications in which AI-powered smart objects can effectively and pervasively run in real-time distributed ML algorithms, shifting part of the raw data analysis and processing from cloud or edge to the device itself. In such context, Artificial Intelligence (AI) can be considered as the backbone of the next generation of Internet of the Things (IoT) devices, which will no longer merely be data collectors and forwarders, but really “smart” devices with built-in data wrangling and data analysis features that leverage lightweight machine learning algorithms to make autonomous decisions on the field. This work thoroughly reviews and analyses the most popular ML algorithms, with particular emphasis on those that are more suitable to run on resource-constrained embedded devices. In addition, several machine learning algorithms have been built on top of a custom multi-dimensional array library. The designed framework has been evaluated and its performance stressed on Raspberry Pi III- and IV-embedded computers.
... In this context, the delay of the data transfer and the availability of connections make cloud unusable when real-time operation is required [9], let alone when privacy is at stake. Instead, processing "at the edge", namely locally on the device and close to the sensors is highly desirable [10]. To enable this, many reduced-precision DL models have been proposed, such as Binarized Neural Networks (BNNs) [11] and Ternary Neural Networks (TNNs) [12], which use only 1 or 2 bits, respectively, to encode the network parameters. ...
... This is shown in [9], where a light harvester has been used to improve the lifetime of a sensor-rich wearable node, absorbing a mean current of 1.75 mA to run a Convolutional Neural Network (CNN). In this context, deep edge computing appears the most viable solution [10] to implement custom processing nodes with a good trade-off between energy efficiency and a sufficient processing power to achieve a high level of recognition accuracy. To this purpose, a new accelerometer/gyroscope MEMS has been presented [11], where simple decision trees are used to embed HAR functionalities. ...
... Wu et al. [4] claimed that user experience can be improved by bringing machine learning inference to the edge, which reduces latency and makes the device less dependent on network connectivity. Zou et al. [5] observed that in emerging applications such as autonomous systems, human-machine interactions, and the Internet of Things (IoT), it is advantageous to process data near or at the source of the data to improve energy and spectrum efficiency and security, and to decrease latency. Hu et al. [6] explored how computational offloading to the edge improves latency and energy consumption relative to the cloud. ...
Article
Full-text available
The mobile terminals used in the logistics industry can be exposed to wildly varying environments, which may hinder effective operation. In particular, those used in cold storages can be subject to frosting in the scanner window when they are carried out of the warehouses to a room-temperature space outside. To prevent this, they usually employ a film heater on the scanner window. However, the temperature and humidity conditions of the surrounding environment and the temperature of the terminal itself that cause frosting vary widely. Due to the complicated frost-forming conditions, existing industrial mobile terminals choose to implement rather simple rules that operate the film heater well above the freezing point, which inevitably leads to inefficient energy use. This paper demonstrates that to avoid such waste, on-device artificial intelligence (AI) a.k.a. edge AI can be readily employed to industrial mobile terminals and can improve their energy efficiency. We propose an artificial-intelligence-based approach that utilizes deep learning technology to avoid the energy-wasting defrosting operations. By combining the traditional temperature-sensing logic with a convolutional neural network (CNN) classifier that visually checks for frost, we can more precisely control the defrosting operation. We embed the CNN classifier in the device and demonstrate that the approach significantly reduces the energy consumption. On our test terminal, the net ratio of the energy consumption by the existing system to that of the edge AI for the heating film is almost 14:1. Even with the common current-dissipation accounted for, our edge AI system would increase the operating hours by 86%, or by more than 6 h compared with the system without the edge AI.
... This solution is also adopted by autonomous vehicle infrastructure, where delays can be disastrous. As presented in Zou et al. (2019), a lot of research has been conducted towards Edge and Fog side computing, following the enormous expansion of the IoT device, and the application of ML mechanisms to process the large amount of transferred data. To this end, the aforementioned study summarizes a number of low-power, State-of-the-art Machine Learning dedicated processors. ...
Article
Full-text available
Recent IoT proliferation has undeniably affected the way organizational activities and business procedures take place within several IoT domains such as smart manufacturing, food supply chain, intelligent transportation systems, medical care infrastructures etc. The number of the interconnected edge devices has dramatically increased, creating a huge volume of transferred data susceptible to leakage, modification or disruption, ultimately affecting the security level, robustness and QoS of the attacked IoT ecosystem. In an attempt to prevent or mitigate network abnormalities while accommodating the cohesiveness among the involved entities, modeling their interrelations and incorporating their structural, content and temporal attributes, graph-based anomaly detection solutions have been repeatedly adopted. In this article we propose, a multi-agent system, with each agent implementing a Graph Neural Network, in order to exploit the collaborative and cooperative nature of intelligent agents for anomaly detection. To this end, against the propagating nature of cyber-attacks such as the Distributed Denial-of-Service (DDoS), we propose a distributed detection scheme, which aims to monitor efficiently the entire network infrastructure. To fulfill this task, we consider employing monitors on active network nodes such as IoT devices, SDN forwarders, Fog Nodes, achieving localization of anomaly detection, distribution of allocated resources such as the bandwidth and power consumption and higher accuracy results. In order to facilitate the training, testing and evaluation activities of the Graph Neural Network algorithm, we create simulated datasets of network flows of various normal and abnormal distributions, out of which we extract essential structural and content features to be passed to neighbouring agents.
... Microservices are services with specific purposes, developed and deployed independently, allowing an easier provision and management of systems that require high concurrency, high availability, high scalability, and low coupling (Liu et al., 2020). Many application providers deploy their microservices in cloud servers, but the deployment possibilities are growing with the increasing adoption of fog and edge computing (Zou et al., 2019;Sunyaev, 2020). Besides, related to the edge computing concept, there is still the multi-access edge computing (MEC, previously called mobile edge computing) (Filali et al., 2020), which enables the cloud/edge benefits for telecommunication operators. ...
Conference Paper
Full-text available
The increasing integration of 5G, multi-access edge computing (MEC), and microservices, benefits the development of applications that demand low coupling, low communication latency, high scalability, and high availability. A usual scenario that deals with such requirements is a video application, either to process inference on video images or process video analytics. Given that video data are considered heavy to process and transmit, we should investigate the best way to handle such data. This work presents an experimental setup for the comparison between four data formats used to send video frames among distributed application components in a MEC server. We measured and analyzed the communication latency when sending video data between distributed parties, considering three scenarios.
... In a more general study, Zou et al. [51] list various hardware technologies that enable or accelerate specific types of AI in the edge. Most of these are designed for CNNs or deep learning in general, but some are aimed at Support Vector Machines (SVM). ...
Article
Full-text available
The use of AI in Smart applications and in the organization of the network edge presents a rapidly advancing research field, with a great variety of challenges and opportunities. This article aims to provide a holistic review of studies from 2019 to 2021 related to the Intelligent Edge, a concept comprising both the use of AI to organize edge networks (Edge Intelligence) and Smart applications in the edge. An introduction is given to the technologies required to understand the state of the art of AI in edge networks, and a taxonomy is provided with “Enabling Technology” for Edge Intelligence“, Organization” of the edge using AI, and AI “Applications” in the edge as its main topics. Research trend data from 2015 to 2020 is presented for various subdivisions of these topics, showing both absolute and relative research interest in each subtopic. The “Organization” aspect, being the main focus of this article, has a more fine-grained subdivision, explaining all contributing factors in detail. The trends indicate an exponential increase in research interest in nearly all subtopics, but significant differences between them. For each subdivision of the taxonomy a number of selected studies from 2019 to 2021 are gathered to form a high-level illustration of the state of the art of Edge Intelligence. From these selected studies and the trend data, a number of short-term challenges and high-level visions for Edge Intelligence are formulated, providing a basis for future work.
... Various applications like smart home, smart city applications using artificial intelligence combined with machine learning to process the data real time as well as discover valuable knowledge and prediction [80]. In the area of IoT data are relayed to fog and edge computing, there also several usages of AI is shown by the authors in [81]. ...
... AI has begun to see the light of the day with automation and implementation occurring at a large scale and fast pace. Likewise, intense research efforts are underway for integrating fog and edge computing with artificial intelligence to enhance the overall performance, including resource, energy management, security, and reliability [154][155][156]. ...
Article
Full-text available
Cloud computing has become integral lately due to the ever-expanding Internet-of-things (IoT) network. It still is and continues to be the best practice for implementing complex computational applications, emphasizing the massive processing of data. However, the cloud falls short due to the critical constraints of novel IoT applications generating vast data, which entails a swift response time with improved privacy. The newest drift is moving computational and storage resources to the edge of the network, involving a decentralized distributed architecture. The data processing and analytics perform at proximity to end-users, and overcome the bottleneck of cloud computing. The trend of deploying machine learning (ML) at the network edge to enhance computing applications and services has gained momentum lately, specifically to reduce latency and energy consumed while optimizing the security and management of resources. There is a need for rigorous research efforts oriented towards developing and implementing machine learning algorithms that deliver the best results in terms of speed, accuracy, storage, and security, with low power consumption. This extensive survey presented on the prominent computing paradigms in practice highlights the latest innovations resulting from the fusion between ML and the evolving computing paradigms and discusses the underlying open research challenges and future prospects.
... e temperature, humidity, lighting, security, audio, and video in the residence are combined into a comprehensive platform for people to alter in order to improve their quality of life in this residential platform, which uses wireless network technology. In order to get a safer and more comfortable home life and to cause as little pollution to the environment as possible, the IoT-based smart home provides a good choice and it is gradually integrating into people's lives [14]. Smart home devices, as shown in Figure 2, touch all aspects of family life and are the most basic IoT applications in the house. ...
Article
Full-text available
The Internet of Things (IoT) wireless network technology is gradually making its way into campus life in today’s information-driven modern world, bringing convenience to students’ daily lives and studies while also placing new demands on the quality of teaching and management effectiveness of universities. In order to make colleges and universities have a more rational approach to talent training, relevant workers need to grasp the characteristics of the data era, keep up with excellent educational examples, innovate educational and teaching methods, and guide students’ learning and growth in the data era. To that end, this paper begins by examining the practicality and importance of IoT for teaching and management in colleges and universities by examining the current state of teaching in colleges and universities, the application of IoT in colleges and universities, and the shortcomings of IoT, as well as the use of network technology on campus. Furthermore, it proposes the IoT education management platform combined with machine learning technology. Finally, we offer an IoT education management platform that combines machine learning technologies, which we believe will be useful to relevant practitioners.
... Despite the extensive researches and numerous experiments on the utilization of AI in optimizing IoV [18][19][20][21][22] , the systematic summaries that identify the basic concepts and development strategies in AI for edge service optimization in IoV remain limited. The existing surveys on IoV mainly tackle the aspects of architectures and applications [23][24][25] , scalability perspectives and quality of services [26] , privacy and security [27] , and the challenges and opportunities [28] . ...
Article
The Internet of Vehicles (IoV) plays a crucial role in providing diversified services because of its powerful capability of collecting real-time information. Generally, collected information is transmitted to a centralized resource-intensive cloud platform for service implementation. Edge Computing (EC) that deploys physical resources near road-side units is involved in IoV to support real-time services for vehicular users. Additionally, many measures are adopted to optimize the performance of EC-enabled IoV, but they hardly help make dynamic decisions according to real-time requests. Artificial Intelligence (AI) is capable of enhancing the learning capacity of edge devices and thus assists in allocating resources dynamically. Although extensive research has employed AI to optimize EC performance, summaries with relative concepts or prospects are quite few. To address this gap, we conduct an exhaustive survey about utilizing AI in edge service optimization in IoV. Firstly, we establish the general condition and relative concepts about IoV, EC, and AI. Secondly, we review the edge service frameworks for IoV and explore the use of AI in edge server placement and service offloading. Finally, we discuss a number of open issues in optimizing edge services with AI.
Article
It remains challenging for a Convolutional Neural Network (CNN) accelerator to maintain high hardware utilization and low processing latency with restricted on-chip memory. This paper presents an In-Execution Configuration Accelerator (IECA) that realizes an efficient control scheme, exploring architectural data reuse, unified in-execution controlling, and pipelined latency hiding to minimize configuration overhead out of the computation scope. The proposed IECA achieves row-wise convolution with tiny distributed buffers and reduces the size of total on-chip memory by removing 40% of redundant memory storage with shared delay chains. By exploiting a reconfigurable Sequence Mapping Table (SMT) and Finite State Machine (FSM) control, the chip realizes cycle-accurate Processing Element (PE) control, automatic loop tiling and latency hiding without extra time slots for pre-configuration. Evaluated on AlexNet and VGG-16, the IECA retains over 97.3% PE utilization and over 95.6% memory access time hiding on average. The chip is designed and fabricated in a UMC 55-nm process running at a frequency of 250 MHz and achieves an area efficiency of 30.55 GOPS/mm² and 0.244 GOPS/KGE (kilo-gate-equivalent), which makes an over 2.0x and 2.1x improvement, respectively, compared with that of previous related works. Implementation of the IEC control scheme uses only a 0.55% area of the 2.75 mm² core.
Article
In this work, a new custom design of an anomaly detection and classification system is proposed. It is composed of a convolutional Auto-Encoder (AE) hardware design to perform anomaly detection which cooperates with a mixed HW/SW Convolutional Neural Network (CNN) to perform the classification of detected anomalies. The AE features a partial binarization, so that the weights are binarized while the activations, associated to some selected layers, are non-binarized. This has been necessary to meet the severe area and energy constraints that allow it to be integrated on the same die as the MEMS sensors for which it serves as a neural accelerator. The CNN shares the feature extraction module with the AE, whereas a SW classifier is triggered by the AE when a fault is detected, working asynchronously to it. The AE has been mapped on a Xilinx Artix-7 FPGA, featuring an Output Data Rate (ODR) of 365 kHz and achieving a power dissipation of $333 \boldsymbol {\mu } $ W/MHz. Logic synthesis has targeted TSMC CMOS 65 nm, 90 nm, and 130 nm standard cells. Best results achieved highlight a power consumption of $138 \boldsymbol {\mu } $ W/MHz with an area occupation of 0.49 mm2 when real-time operations are set. These results enable the integration of the complete neural accelerator in the CMOS circuitry that typically sits with the inertial MEMS on the same silicon die. Comparisons with the related works suggest that the proposed system is capable of state-of-the-art performances and accuracy.
Article
This study describes a line cloud architecture and IDM-based energy metering system to replace existing meter reading methods. They can regularly monitor meter readings without sending someone to each residence, and the bill is automatically sent to each user via IDM. If the consumer fails to pay the bill, the service provider can cut off the supply; this technology will prevent the illicit use of electricity, often known as power detection, and locate line faults rapidly and precisely without the need for human intervention. Because of the increased deployment of energy meters, much data on electric energy is used. Developing cloud architecture technologies could utilize this data better to prevent power detection. In this paper, an intelligent detection method of electric energy meter overload functioning state based on cloud architecture (IDM EEMOF-CA) is provided in detail and utilized to identify electric energy detection. There are currently no studies involving the use of IDM EEM-CA to detect power exposure to the authors' knowledge.
Article
Full-text available
Distributed edge intelligence is a disruptive research area that enables the execution of machine learning and deep learning (ML/DL) algorithms close to where data are generated. Since edge devices are more limited and heterogeneous than typical cloud devices, many hindrances have to be overcome to fully extract the potential benefits of such an approach (such as data-in-motion analytics). In this paper, we investigate the challenges of running ML/DL on edge devices in a distributed way, paying special attention to how techniques are adapted or designed to execute on these restricted devices. The techniques under discussion pervade the processes of caching, training, inference, and offloading on edge devices. We also explore the benefits and drawbacks of these strategies.
Article
A custom Human Activity Recognition system is presented based on the resource-constrained Hardware (HW) implementation of a new partially binarized Hybrid Neural Network. The system processes data in real-time from a single tri-axial accelerometer, and is able to classify between 5 different human activities with an accuracy of 97.5% when the Output Data Rate of the sensor is set to 25 Hz. The new Hybrid Neural Network (HNN) has binary weights (i.e. constrained to +1 or −1) but uses non-binarized activations for some layers. This, in conjunction with a custom pre-processing module, achieves much higher accuracy than Binarized Neural Network. During pre-processing, the measurements are made independent from the spatial orientation of the sensor by exploiting a reference frame transformation. A prototype has been realized in a Xilinx Artix 7 FPGA, and synthesis results have been obtained with TSMC CMOS 65 nm LP HVT and 90 nm standard cells. Best result shows a power consumption of $6.3~\mu \text{W}$ and an area occupation of 0.2 mm $^{\mathbf {2}}$ when real-time operations are set, enabling in this way, the possibility to integrate the entire HW accelerator in the auxiliary circuitry that normally equips inertial Micro Electro-Mechanical Systems (MEMS).
Article
Full-text available
Industry 4.0 has promised to bring new digital ecosystem for various key industrial applications in terms of providing secure, intelligent, autonomous and self-adaptive industrial IoT (IIoT) networks. Such industrial systems will be relatively more complex because of the involvement of the vast number of the heterogeneous sensors/devices and diverse nature of various stakeholders and service providers. These complex industrial processes also have strict requirements in terms of low-latency services/resources together with security, trust and process mon-itoring/tracking among others. In this context, Blockchain and Edge Computing emerge as prominent technologies to strengthen the vision of the digitization for the various industries by fulfilling/addressing the essential requirements for processes. For example, edge paradigm ensures low latency services for IIoT applications and on the other hand, blockchain provides key features such as security, trust and decentralization among others. In this paper, we have proposed a 'BlockEdge' framework that primarily combines these two enabling technologies to address some of the critical issues faced by the current IIoT networks. We have also provided performance evaluation of the proposed framework using well-known simulation environment for edge/fog computing, i.e.'iFogSim' and compared with the existing IIoT model which does not include blockchain.
Article
Full-text available
Solid waste management (SWM) is a crucial management entity in urban cities to handle the waste from its generation to disposal to accomplish a clean environment. The waste management operation mainly encompasses various climatic, demographic, environmental, legislative, technological, and socioeconomic dimensions. The traditional approaches deliver limitations in the process of predict and optimizing such composite non‐linear operations. The integration of the internet of things (IoT) and artificial intelligence (AI) methods have progressively gained attention by delivering potential alternatives for resolving the difficulties in SWM. This article presents a review of the significance of the amalgamation of IoT and machine learning (ML) in the SWM to predict waste generation, waste classification, route optimization, estimation of methane emissions, and so forth. The article covers the application of each ML model for the activities, including SWM, compositing, incineration, pyrolysis, gasification, landfill, and anaerobic digestion. Moreover, it is concluded that the decision tree and random forest (DT‐RF) algorithm is minor implemented, and artificial neural network (ANN) is implemented majorly in the SWM. The large number of data sets covered in the publication are secured and hidden; it limits replicating the AI models; this is also one key constraint of the non‐implementation of AI models in SWM. Scarcity of data, accurate data, rare availability of customized AI models for tackling the activities in SWM are the limitations identified from the previous studies. Implementation of low‐power ML processors, edge and fog computing‐based devices is the future direction for overcoming SWM limitations.
Article
With the advent of Internet of Things (IoT) connecting billions of mobile and stationary devices to serve real-time applications, cloud computing paradigms face some significant challenges such as high latency and jitter, non-supportive location-awareness and mobility, and non-adaptive communication types. To address these challenges, edge computing paradigms, namely Fog Computing (FC), Mobile Edge Computing (MEC) and Cloudlet, have emerged to shift the digital services from centralized cloud computing to computing at edges. In this article, we analyse cloud and edge computing paradigms from features and pillars perspectives to identify the key motivators of the transitions from one type of virtualized computing paradigm to another one. We then focus on computing and network virtualization techniques as the essence of all these paradigms, and delineate why virtualization features, resource richness and application requirements are the primary factors for the selection of virtualization types in IoT frameworks. Based on these features, we compare the state-of-the-art research studies in the IoT domain. We finally investigate the deployment of virtualized computing and networking resources from performance perspective in an edge-cloud environment, followed by mapping of the existing work to the provided taxonomy for this research domain. The lessons from the reviewed are that the selection of virtualization technique, placement and migration of virtualized resources rely on the requirements of IoT services (i.e., latency, scalability, mobility, multi-tenancy, privacy, and security). As a result, there is a need for prioritizing the requirements, integrating different virtualization techniques, and exploiting a hierarchical edge-cloud architecture.
Article
The impact that artificial intelligence (AI) has made across several industries in today’s society is clearly seen in applications ranging from medical diagnosis to customer service chatbots, to financial trading. It is also evident that AI has a huge role to play in emerging and future applications and will be increasingly used in mission-critical and time-sensitive applications such as remote surgeries, cybersecurity and self-driving cars. To satisfy the latency, security and privacy requirements that such applications require, AI which has gained its merit by utilising resource-heavy cloud infrastructure, needs to perform well in resource-constrained environments at the network edge. To address this need, this paper characterises the cloud-to-thing continuum and provides an architecture for enabling AI in fully edge-based scenarios. In addition, the paper provides strategies to tackle the communication inefficiencies that arise from the distributed nature of fully edge-based scenarios. Performance improvements exhibited by these strategies in state-of-the art research is presented, as well as directions where further advancements can be made. The material is presented in a simple manner to catalyse the understanding and hence the participation of multidisciplinary researchers in addressing this challenge.
Article
Solid state ionics has grown to be one of the most important directions of science, combining emerging interdisciplinary technologies for many applications. This paper begins with the brief history of solid state ionics, followed by detailed discussions on scientific problems, state-of-art theoretical and experimental methods, and promising research directions in this field. The practical applications of principles and knowledge in solid state ionics are also summarized.
Article
Machine Learning (ML) techniques have changed the analysis of massive data in the Internet of Things (IoT) environment very effectively. In the IoT theme of applications, reducing latency and energy consumption are the two crucial network Quality of Service (QoS) parameters and the most significant challenges because they directly impact the users’ experience. Enabling intelligence at the IoT fog computing framework with ML classifiers' help determines the computing requirements that, in turn, help to execute the vast data collected in the IoT fog computing for real-time operations efficiently. In this paper, the exploration of ML algorithms on the resource constraint IoT fog computing framework and the determination of the suitable ML classifier for reducing latency and energy levels with the usage of ambient sensors in the IoT theme are presented.
Article
Full-text available
AIoT (artificial intelligence of things) is an integrated product of artificial intelligence (AI) and internet of things (IoT).Currently, it has been widely used in smart cities, smart homes, smart manufacturing, and driverless.However, the research of AIoT is still in its infancy, facing with many problems and challenges.In order to clarify the concept and provide possible solutions, a comprehensive survey was carried out on AIoT.Firstly, a clear definition of AIoT was provided, along with the brief introduction on its background and application scenarios.And then a novel cloud-edge-end hybrid AIoT architecture for intelligent information processing was constructed.Based on the research framework of AIoT, the research status and solutions were discussed, including AI integrated IoT data acquisition, complex event processing and coordination, cloud-edge-end integration, AI-enhanced IoT security and privacy, and AI-based applications, etc.Finally, it identified the open challenges and offers future research directions.
Chapter
Internet of Things (IoT) is an interconnected wireless network where smart nodes (IoT devices) interact with each other in order to exchange data through the communicating medium. IoT technology has become important for people to build smart systems upon the use of technology. Internet of things is realized by the idea of free flow of information among various low-power embedded devices that use the Internet to communicate with one another. In the recent past IoT have grown rapidly and have become an extension of existing universal Internet. It can be easily anticipated that large scale systems equipped with numerous sensors will prevail in our society. With the rise of the Internet of Things (IoT) technology, the number of IoT devices/sensors has also increased significantly Billions of smart devices connected to IoT environment can communicate among themselves using sensors and actuators. This rapid growth and inclusion of IoT technologies in our daily life are facing challenges since most of the devices specially sensors in IoT network are resource constraint in terms of energy, computation capability, etc. Data collected from these sensors sent through the middleware like gateway, routers, etc. to the cloud servers or toward various analytical engine for meaningful knowledge discovery. These processed data and knowledge have lately attracted huge attention, and organizations are excited about the business value of the data that will be generated by deploying such networks. With this advent of IoT it has also attracted various security and privacy concerns. Due to the structurally open IoT architecture and the tremendous usage of the paradigm itself, causes to generate many unconventional security issues for the existing networking technologies. Moreover, since sensor nodes are cooperative within the IoT network, this sharing of data can create new challenges that can disrupt the systems’ regular functionalities and operations. In another aspect the growth of IoT technologies has enhanced by assimilating them with cloud computing and the era IoT-cloud has emerged. With these, some new class of security and privacy issues have also introduced. Furthermore, the commercialization of the IoT has led to several public security concerns including threats of cyber-attacks, privacy issues, and fraud crimes. This chapter gathers the needed information to give a complete picture of security issues and problems faced in IoT communication. In this chapter, we detail the major security as well as privacy issues more specifically. An extensive description of security threats and challenges across the different layers of the architecture of IoT systems is represented. The issues related to IoT-cloud is also highlighted. The light will be shed on the state-of-the-art solutions to the emerging and latest security issues in this field. We hereby present the evolving resolve policies as mined from the research work of various authors in this field that will expose the several research areas in IoT-cloud era. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
Article
Full-text available
Tensor processing units improve performance per watt of neural networks in Google datacenters by roughly 50x.
Article
Full-text available
A wide range of Internet of Things devices, platforms and applications have been implemented in the past decade. The variation in platforms, communication protocols and data formats of these systems creates islands of applications. Many organizations are working towards standardizing the technologies used at different layers of communication in these systems. However, interoperability still remains one of the main challenges towards realizing the grand vision of IoT. Intergration approaches proven in the existing Internet or enterprise applications are not suitable for the IoT, mainly due to the nature of the devices involved; the majority of the devices are resource constrained. To address this problem of interoperability, our work considers various types of IoT application domains, architecture of the IoT and the works of standards organizations to give a holistic abstract model of IoT. According to this model, there are three computing layers, each with a different level of interoperability needs — technical, syntactic or semantic. This work presents a Web of Virtual Things (WoVT) server that can be deployed at the middle layer of IoT (Fog layer) and Cloud to address the problem of interoperability. It exposes a REST like uniform interface for syntactic integration of devices at the bottom layer of IoT (perception layer). An additional RESTful api is used for integration with other similar WoVT servers at the Fog or the Cloud layer. The server uses a state of the art architecture to enable this integration pattern and provides means towards semantic interoperability. The analysis and evaluation of the implementation, such as performance, resource utilization and security perspectives, are presented. The simulation results demonstrate that an integrated and scalable IoT through the web of virtual things can be realized.
Article
Full-text available
Digitalization is a global trend becoming ever more important to our connected and sustainable society. This trend also affects industry where the Industrial Internet of Things is an important part, and there is a need to conserve spectrum as well as energy when communicating data to a fog or cloud back-end system. In this paper we investigate the benefits of fog computing by proposing a novel distributed learning model on the sensor device and simulating the data stream in the fog, instead of transmitting all raw sensor values to the cloud back-end. To save energy and to communicate as few packets as possible, the updated parameters of the learned model at the sensor device are communicated in longer time intervals to a fog computing system. The proposed framework is implemented and tested in a real world testbed in order to make quantitative measurements and evaluate the system. Our results show that the proposed model can achieve a 98% decrease in the number of packets sent over the wireless link, and the fog node can still simulate the data stream with an acceptable accuracy of 97%. We also observe an end-to-end delay of 180 ms in our proposed three-layer framework. Hence, the framework shows that a combination of fog and cloud computing with a distributed data modeling at the sensor device for wireless sensor networks can be beneficial for Industrial Internet of Things applications.
Article
Full-text available
A versatile reconfigurable accelerator architecture for binary/ternary deep neural networks is presented. In-memory neural network processing without any external data accesses, sustained by the symmetry and simplicity of the computation of the binary/ternaty neural network, improves the energy efficiency dramatically. The prototype chip is fabricated, and it achieves 1.4 TOPS (tera operations per second) peak performance with 0.6-W power consumption at 400-MHz clock. The application examination is also conducted.
Article
Full-text available
Rapid developments in hardware, software, and communication technologies have allowed the emergence of Internet-connected sensory devices that provide observation and data measurement from the physical world. By 2020, it is estimated that the total number of Internet-connected devices being used will be between 25 and 50 billion. As the numbers grow and technologies become more mature, the volume of data published will increase. Internet-connected devices technology, referred to as Internet of Things (IoT), continues to extend the current Internet by providing connectivity and interaction between the physical and cyber worlds. In addition to increased volume, the IoT generates Big Data characterized by velocity in terms of time and location dependency, with a variety of multiple modalities and varying data quality. Intelligent processing and analysis of this Big Data is the key to developing smart IoT applications. This article assesses the different machine learning methods that deal with the challenges in IoT data by considering smart cities as the main use case. The key contribution of this study is presentation of a taxonomy of machine learning algorithms explaining how different techniques are applied to the data in order to extract higher level information. The potential and challenges of machine learning for IoT data analytics will also be discussed. A use case of applying Support Vector Machine (SVM) on Aarhus Smart City traffic data is presented for a more detailed exploration.
Article
Full-text available
The evolution of the Internet of things (IoT) and the continuing increase in the number of sensors connected to the Internet impose big challenges regarding the management of the resulting deluge of data and network latency. Uploading sensor data over the web does not add value. Therefore, an efficient knowledge extraction technique is badly needed to reduce the amount of data transfer and to help simplify the process of knowledge management. Homoscedasticity and statistical features extraction are introduced in this paper as novelty detection enabling techniques, which help extract the important events in sensor data in real time when used with neural classifiers. Experiments have been conducted on a fog computing platform. System performance has been also evaluated on an occupancy dataset and showed promising results.
Article
Full-text available
The increasing use of wearables in smart telehealth system led to generation of huge medical big data. Cloud and fog services leverage these data for assisting clinical procedures.IoT Healthcare has been benefited from this large pool of generated data. This paper suggests the use of low-resource machine learning on Fog devices kept close to wearables for smart telehealth. For traditional telecare systems, the signal processing and machine learning modules are deployed in cloud that process physiological data. This paper presents a Fog architecture that relied on unsupervised machine learning big data analysis for discovering patterns in physiological data. We developed a prototype using Intel Edison and Raspberry Pi that was tested on real-world pathological speech data from telemonitoring of patients with Parkinson's disease (PD). Proposed architecture employed machine learning for analysis of pathological speech data obtained from smartwatches worn by the patients with PD. Results show that proposed architecture is promising for low-resource machine learning. It could be useful for other applications within wearable IoT for smart telehealth scenarios by translating machine learning approaches from the cloud back-end to edge computing devices such as Fog. Keywords: Dysarthria; Edge Computing; Fog Computing; K-means Clustering, Parkinson's Disease; Speech Disorders
Conference Paper
Full-text available
Computational offloading services at the edge of the Internet for mobile devices are becoming a reality. Using a wide range of mobile applications, we explore how such infrastructure improves latency and energy consumption relative to the cloud. We present experimental results from WiFi and 4G LTE networks that confirm substantial wins from edge computing for highly interactive mobile applications.
Article
Full-text available
Many cloud-based applications employ a data centre as a central server to process data that is generated by edge devices, such as smartphones, tablets and wearables. This model places ever increasing demands on communication and computational infrastructure with inevitable adverse effect on Quality-of-Service and Experience. The concept of Edge Computing is predicated on moving some of this computational load towards the edge of the network to harness computational capabilities that are currently untapped in edge nodes, such as base stations, routers and switches. This position paper considers the challenges and opportunities that arise out of this new direction in the computing landscape.
Conference Paper
Full-text available
Many cloud-based applications employ a data centre as a central server to process data that is generated by edge devices, such as smartphones, tablets and wearables. This model places ever increasing demands on communication and computational infrastructure with inevitable adverse effect on Quality-of-Service and Experience. The concept of Edge Computing is predicated on moving some of this computational load towards the edge of the network to harness computational capabilities that are currently untapped in edge nodes, such as base stations, routers and switches. This position paper considers the challenges and opportunities that arise out of this new direction in the computing landscape.
Conference Paper
Full-text available
Despite the increasing usage of cloud computing, there are still issues unsolved due to inherent problems of cloud computing such as unreliable latency, lack of mobility support and location-awareness. Fog computing can address those problems by providing elastic resources and services to end users at the edge of network, while cloud computing are more about providing resources distributed in the core network. This survey discusses the definition of fog computing and similar concepts, introduces representative application scenarios, and identifies various aspects of issues we may encounter when designing and implementing fog computing systems. It also highlights some opportunities and challenges, as direction of potential future work, in related techniques that need to be considered in the context of fog computing.
Conference Paper
Full-text available
Deep learning using convolutional neural networks (CNN) gives state-of-the-art accuracy on many computer vision tasks (e.g. object detection, recognition, segmentation). Convolutions account for over 90% of the processing in CNNs for both inference/testing and training, and fully convolutional networks are increasingly being used. To achieve state-of-the-art accuracy requires CNNs with not only a larger number of layers, but also millions of filters weights, and varying shapes (i.e. filter sizes, number of filters, number of channels) as shown in Fig. 14.5.1. For instance, AlexNet [1] uses 2.3 million weights (4.6MB of storage) and requires 666 million MACs per 227×227 image (13kMACs/pixel). VGG16 [2] uses 14.7 million weights (29.4MB of storage) and requires 15.3 billion MACs per 224×224 image (306kMACs/pixel). The large number of filter weights and channels results in substantial data movement, which consumes significant energy.
Article
Convolutional neural networks (CNN) provide state-of-the-art results in a wide variety of machine learning (ML) applications, ranging from image classification to speech recognition. However, they are very computationally intensive and require huge amounts of storage. Recent work strived towards reducing the size of the CNNs: [1] proposes a binary-weight-network (BWN), where the filter weights (w i 's) are ±1 (with a common scaling factor per filter: α). This leads to a significant reduction in the amount of storage required for the W i 's, making it possible to store them entirely on-chip. However, in a conventional all-digital implementation [2, 3], reading the wj i s and the partial sums from the embedded SRAMs require a lot of data movement per computation, which is energy-hungry. To reduce data-movement, and associated energy, we present an SRAM-embedded convolution architecture (Fig. 31.1.1), which does not require reading the w i 's explicitly from the memory. Prior work on embedded ML classifiers have focused on 1b outputs [4] or a small number of output classes [5], both of which are not sufficient for CNNs. This work uses 7b inputs/outputs, which is sufficient to maintain good accuracy for most of the popular CNNs [1]. The convolution operation is implemented as voltage averaging (Fig. 31.1.1), since the wj's are binary, while the averaging factor (1/N) implements the weight-coefficient α (with a new scaling factor, M, implemented off-chip).
Article
QUEST is a programmable multiple instruction, multiple data (MIMD) parallel accelerator for general-purpose state-of-the-art deep neural networks (DNNs). It features die-to-die stacking with three-cycle latency, 28.8 GB/s, 96 MB, and eight SRAMs using an inductive coupling technology called the ThruChip interface (TCI). By stacking the SRAMs instead of DRAMs, lower memory access latency and simpler hardware are expected. This facilitates in balancing the memory capacity, latency, and bandwidth, all of which are in demand by cutting-edge DNNs at a high level. QUEST also introduces log-quantized programmable bit-precision processing for achieving faster (larger) DNN computation (size) in a 3-D module. It can sustain higher recognition accuracy at a lower bitwidth region compared to linear quantization. The prototype QUEST chip is integrated in the 40-nm CMOS technology, and it achieves 7.49 tera operations per second (TOPS) peak performance in binary precision, and 1.96 TOPS in 4-bit precision at 300-MHz clock.
Article
The trend of pushing inference from cloud to edge due to concerns of latency, bandwidth, and privacy has created demand for energy-efficient neural network hardware. This paper presents a mixed-signal binary convolutional neural network (CNN) processor for always-on inference applications that achieves 3.8 μJ/classification at 86% accuracy on the CIFAR-10 image classification data set. The goal of this paper is to establish the minimum-energy point for the representative CIFAR-10 inference task, using the available design tradeoffs. The BinaryNet algorithm for training neural networks with weights and activations constrained to +1 and -1 drastically simplifies multiplications to XNOR and allows integrating all memory on-chip. A weight-stationary, data-parallel architecture with input reuse amortizes memory access across many computations, leaving wide vector summation as the remaining energy bottleneck. This design features an energy-efficient switched-capacitor (SC) neuron that addresses this challenge, employing a 1024-bit thermometer-coded capacitive digital-to-analog converter (CDAC) section for summing pointwise products of CNN filter weights and activations and a 9-bit binary-weighted section for adding the filter bias. The design occupies 6 mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> in 28-nm CMOS, contains 328 kB of on-chip SRAM, operates at 237 frames/s (FPS), and consumes 0.9 mW from 0.6 V/0.8 V supplies. The corresponding energy per classification (3.8 μJ) amounts to a 40× improvement over the previous low-energy benchmark on CIFAR-10, achieved in part by sacrificing some programmability. The SC neuron array is 12.9× more energy efficient than a synthesized digital implementation, which amounts to a 4× advantage in system-level energy per classification.
Chapter
The development of printed electronics provides new possibilities for the cost‐effective manufacture of large‐area flexible devices. The inkjet printing of silver interconnections on flexible substrates is suitable for printed electronic systems. This chapter presents a heterogeneous integration platform enabled by inkjet printing technology. The integration process shows great potential for the implementation of heterogeneous systems. The heterogeneous system is in strong demand, to combine high‐performance silicon electronics, which is of low cost per function, and flexible printed electronics, which is of low cost per area. The bending test indicates that the heterogeneous integrated system can work reliably for several applications such as healthcare devices and intelligent packaging. In order to improve the reliability and protect the conductive circuits against the environment, a thin layer of dielectric material could be added on top of the printed electronics as insulation and protection.
Article
Sustained energy efficiency improvements have been instrumental for the vertiginous evolution of electronic systems with computing, sensing, communication and storage capabilities. Energy efficiency improvements are indeed crucial for continued increase in the performance under a limited power budget, reduced operating cost, as well as for untethering traditionally wired systems. This is indeed true for high-performance systems subject to heat removal limitations (e.g., server blades), as well as for operational cost considerations when the cost of electricity is a major fraction of the total cost, as in the case of datacenters [1] , or the more recent crypto-currency mining endeavors [2] . Energy reductions are also critical in portable electronics, due to the limited thermal budget and battery energy availability. Similarly, energy reductions are essential in miniaturized energy-autonomous systems such as sensor nodes, hearables, wearables and others, due to their tightly constrained energy source [3] . Overall, energy efficiency improvements have historically permitted the continuous size down-scaling and lifetime extension of electronic systems (see, [4] ).
Article
This paper presents a 28-nm system-on-chip (SoC) for Internet of things (IoT) applications with a programmable accelerator design that implements a powerful fully connected deep neural network (DNN) classifier. To reach the required low energy consumption, we exploit the key properties of neural network algorithms: parallelism, data reuse, small/sparse data, and noise tolerance. We map the algorithm to a very large scale integration (VLSI) architecture based around an single-instruction, multiple-data data path with hardware support to exploit data sparsity by completely eliding unnecessary computation and data movement. This approach exploits sparsity, without compromising the parallel computation. We also exploit the inherent algorithmic noise-tolerance of neural networks, by introducing circuit-level timing violation detection to allow worst case voltage guard-bands to be minimized. The resulting intermittent timing violations may result in logic errors, which conventionally need to be corrected. However, in lieu of explicit error correction, we cope with this by accentuating the noise tolerance of neural networks. The measured test chip achieves high classification accuracy (98.36% for the MNIST test set), while tolerating aggregate timing violation rates > 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−1</sup> . The accelerator achieves a minimum energy of 0.36 $\mu \text{J}$ /inference at 667 MHz; maximum throughput at 1.2 GHz and 0.57 $\mu \text{J}$ /inference; or a 10% margined operating point at 1 GHz and 0.58 $\mu \text{J}$ /inference.
Conference Paper
Container-based virtualization offers a very feasible alternative to heavyweights like KVM or XEN. Containers are lightweight and offer near-native performance. They are also easy to deploy because of continuous integration/development tools and environments. This paper offers a brief introduction to containers, defines its properties and provides use-cases in the context of those properties. Secondly, we look at the live migration of stateful applications via containers. Live migration is a promising technique at the network edge to unload computing to other nodes to expand the use of fog computing / mobile-edge computing. Our experiment shows that live migration of stateful applications can result in three different types of errors, namely, resend, reprocessed and wrong-order errors.
Article
Fog Computing (FC) is a flexible architecture to support distributed domain-specific applications with cloud-like quality of service. However, current FC still lacks the mobility support mechanism when facing many mobile users with diversified application quality requirements. Such mobility support mechanism can be critical such as in the industrial internet where human, products, and devices are moveable. To fill in such gaps, in this paper we propose novel container migration algorithms and architecture to support mobility tasks with various application requirements. Our algorithms are realized from three aspects: 1) We consider mobile application tasks can be hosted in a container of a corresponding fog node that can be migrated, taking the communication delay and computational power consumption into consideration; 2) We further model such container migration strategy as multiple dimensional Markov Decision Process (MDP) spaces. To effectively reduce the large MDP spaces, efficient deep reinforcement learning algorithms are devised to achieve fast decision-making and 3) We implement the model and algorithms as a container migration prototype system and test its feasibility and performance. Extensive experiments show that our strategy outperforms the existing baseline approaches 2.9%, 48.5% and 58.4% on average in terms of delay, power consumption, and migration cost, respectively.
Article
Hybrid neural networks (hybrid-NNs) have been widely used and brought new challenges to NN processors. Thinker is an energy efficient reconfigurable hybrid-NN processor fabricated in 65-nm technology. To achieve high energy efficiency, three optimization techniques are proposed. First, each processing element (PE) supports bit-width adaptive computing to meet various bit-widths of neural layers, which raises computing throughput by 91% and improves energy efficiency by 1.93x on average. Second, PE array supports on-demand array partitioning and reconfiguration for processing different NNs in parallel, which results in 13.7% improvement of PE utilization and improves energy efficiency by 1.11x. Third, a fused data pattern-based multi-bank memory system is designed to exploit data reuse and guarantee parallel data access, which improves computing throughput and energy efficiency by 1.11x and 1.17x, respectively. Measurement results show that this processor achieves 5.09-TOPS/W energy efficiency at most.
Article
We describe the emerging paradigm of self-aware computing and give an overview of proposed architectures and applications with focus on SoC solutions.
Conference Paper
State-of-the-art deep neural networks (DNNs) have hundreds of millions of connections and are both computationally and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources and power budgets. While custom hardware helps the computation, fetching weights from DRAM is two orders of magnitude more expensive than ALU operations, and dominates the required power. Previously proposed 'Deep Compression' makes it possible to fit large DNNs (AlexNet and VGGNet) fully in on-chip SRAM. This compression is achieved by pruning the redundant connections and having multiple connections share the same weight. We propose an energy efficient inference engine (EIE) that performs inference on this compressed network model and accelerates the resulting sparse matrix-vector multiplication with weight sharing. Going from DRAM to SRAM gives EIE 120x energy saving; Exploiting sparsity saves 10x; Weight sharing gives 8x; Skipping zero activations from ReLU saves another 3x. Evaluated on nine DNN benchmarks, EIE is 189x and 13x faster when compared to CPU and GPU implementations of the same DNN without compression. EIE has a processing power of 102GOPS/s working directly on a compressed network, corresponding to 3TOPS/s on an uncompressed network, and processes FC layers of AlexNet at 1.88x10^4 frames/sec with a power dissipation of only 600mW. It is 24,000x and 3,400x more energy efficient than a CPU and GPU respectively. Compared with DaDianNao, EIE has 2.9x, 19x and 3x better throughput, energy efficiency and area efficiency.
Article
We propose distributed deep neural networks (DDNNs) over distributed computing hierarchies, consisting of the cloud, the edge (fog) and end devices. While being able to accommodate inference of a deep neural network (DNN) in the cloud, a DDNN also allows fast and localized inference using shallow portions of the neural network at the edge and end devices. When supported by a scalable distributed computing hierarchy, a DDNN can scale up in neural network size and scale out in geographical span. Due to its distributed nature, DDNNs enhance sensor fusion, system fault tolerance and data privacy for DNN applications. In implementing a DDNN, we map sections of a DNN onto a distributed computing hierarchy. By jointly training these sections, we minimize communication and resource usage for devices and maximize usefulness of extracted features which are utilized in the cloud. The resulting system has built-in support for automatic sensor fusion and fault tolerance. As a proof of concept, we show a DDNN can exploit geographical diversity of sensors to improve object recognition accuracy and reduce communication cost. In our experiment, compared with the traditional method of offloading raw sensor data to be processed in the cloud, DDNN locally processes most sensor data on end devices while achieving high accuracy and is able to reduce the communication cost by a factor of over 20x.
Article
An energy-efficient restricted Boltzmann machine (RBM) processor (RBM-P) supporting on-chip learning and inference is proposed for machine learning and Internet of Things (IoT) applications in this paper. To train a neural network (NN) model, the RBM structure is applied to supervised and unsupervised learning, and a multi-layer NN can be constructed and initialized by stacking multiple RBMs. Featuring NN model reduction for external memory bandwidth saving, low power neuron binarizer (LPNB) with dynamic clock gating and area-efficient NN-like activation function calculators for power reduction, user-defined connection map (UDCM) for both computation time and bandwidth saving, and early stopping (ES) mechanism for learning process, the proposed system integrates 32 RBM cores with maximal 4k neurons per layer and 128 candidates per sample for machine learning applications. Implemented in 65nm CMOS technology, the proposed RBM-P chip costs 2.2 M gates and 128 kB SRAM with 8.8 mm² area. Operated at 1.2 V and 210 MHz, this chip achieves 7.53G neuron weights (NWs) and 11.63G NWs per second with 41.3 and 26.7 pJ per NW for learning and inference, respectively.
Article
Demand for highly energy-efficient coprocessor for the inference computation of deep neural networks is increasing. We propose the time-domain neural network (TDNN), which employs time-domain analog and digital mixed-signal processing (TDAMS) that uses delay time as the analog signal. TDNN not only exploits energy-efficient analog computing, but also enables fully spatially unrolled architecture by the hardware-efficient feature of TDAMS. The proposed fully spatially unrolled architecture reduces energy-hungry data moving for weight and activations, thus contributing to significant improvement of energy efficiency. We also propose useful training techniques that mitigate the non-ideal effect of analog circuits, which enables to simplify the circuits and leads to maximizing the energy efficiency. The proof-of-concept chip shows unprecedentedly high energy efficiency of 48.2 TSop/s/W.
Conference Paper
With technologies developed in the Internet of Things, embedded devices can be built into every fabric of urban environments and connected to each other; and data continuously produced by these devices can be processed, integrated at different levels, and made available in standard formats through open services. The data, obviously f a form of “big data”, is now seen as the most valuable asset in developing intelligent applications. As the sizes of the IoT data continue to grow, it becomes inefficient to transfer all the raw data to a centralised, cloud-based data centre and to perform efficient analytics even with the state-ofthe- art big data processing technologies. To address the problem, this article demonstrates the idea of “distributed intelligence” for sensor data computing, which disperses intelligent computation to the much smaller while autonomous units, e.g., sensor network gateways, smart phones or edge clouds in order to reduce data sizes and to provide high quality data for data centres. As these autonomous units are usually in close proximity to data consumers, they also provide potential for reduced latency and improved quality of services. We present our research on designing methods and apparatus for distributed computing on sensor data, e.g., acquisition, discovery, and estimation, and provide a case study on urban air pollution monitoring and visualisation.
Conference Paper
Cloud Computing provides us a means to upload data and use applications over the internet. As the number of devices connecting to the cloud grows, there is undue pressure on the cloud infrastructure. Fog computing or Network Based Computing or Edge Computing allows to move a part of the processing in the cloud to the network devices present along the node to the cloud. Therefore the nodes connected to the cloud have a better response time. This paper proposes a method of moving the computation from the cloud to the network by introducing an android like appstore on the networking devices.
Article
The success of the Internet of Things and rich cloud services have helped create the need for edge computing, in which data processing occurs in part at the network edge, rather than completely in the cloud. Edge computing could address concerns such as latency, mobile devices' limited battery life, bandwidth costs, security, and privacy.
Article
State-of-the-art deep neural networks (DNNs) have hundreds of millions of connections and are both computationally and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources and power budgets. While custom hardware helps the computation, fetching weights from DRAM is two orders of magnitude more expensive than ALU operations, and dominates the required power. Previously proposed 'Deep Compression' makes it possible to fit large DNNs (AlexNet and VGGNet) fully in on-chip SRAM. This compression is achieved by pruning the redundant connections and having multiple connections share the same weight. We propose an energy efficient inference engine (EIE) that performs inference on this compressed network model and accelerates the resulting sparse matrix-vector multiplication with weight sharing. Going from DRAM to SRAM gives EIE 120× energy saving; Exploiting sparsity saves 10×; Weight sharing gives 8×; Skipping zero activations from ReLU saves another 3×. Evaluated on nine DNN benchmarks, EIE is 189× and 13× faster when compared to CPU and GPU implementations of the same DNN without compression. EIE has a processing power of 102 GOPS working directly on a compressed network, corresponding to 3 TOPS on an uncompressed network, and processes FC layers of AlexNet at 1.88×10⁴ frames/sec with a power dissipation of only 600mW. It is 24,000× and 3,400× more energy efficient than a CPU and GPU respectively. Compared with DaDianNao, EIE has 2.9×, 19× and 3× better throughput, energy efficiency and area efficiency.
Article
The cloud is migrating to the edge of the network, where routers themselves may become the virtualisation infrastructure, in an evolution labelled as \the fog". However, many other complementary technologies are reaching a high level of maturity. Their interplay may dramatically shift the information and communication technology landscape in the following years, bringing separate technologies into a common ground. This paper o ers a comprehensive definition \the fog", comprehending technologies as diverse as cloud, sensor networks, peer-to-peer networks, network virtualisation functions or configuration management techniques. We highlight the main challenges faced by this potentially break-through technology amalgamation.