High Network Latency Research Articles

Computation Offloading and Band Selection for IoT Devices in Multi-Access Edge Computing

The advent of Multi-Access Edge Computing (MEC) has enabled service providers to mitigate high network latencies often encountered in accessing cloud services. The key idea of MEC involves service providers deploying containerized application services on MEC servers situated near Internet-of-Things (IoT) device users. The users access these services via wireless base stations with ultra low latency. Computation tasks of IoT devices can then either be executed locally on the devices or on the MEC servers. A key cornerstone of the MEC environment is an offloading policy utilized to determine whether to execute computation tasks on IoT devices or to offload the tasks to MEC servers for processing. In this work, we propose a two phase Probabilistic Model Checking based offloading policy catering to IoT device user preferences. The first stage evaluates the trade-offs between local vs server execution while the second stage evaluates the trade-offs between choice of wireless communication bands for offloaded tasks. We present experimental results in practical scenarios on data gathered from an IoT test-bed setup with benchmark applications to show the benefits of an adaptive preference-aware approach over conventional approaches in the MEC offloading context.

Scalable Distributed Inverted List Indexes in Disaggregated Memory

Memory disaggregation separates compute (CPU) and main memory resources into disjoint physical units to enable elastic and independent scaling. Connected via high-speed RDMA-enabled networks, compute nodes can directly access remote memory. This setting often requires complex protocols with many network roundtrips as memory nodes have near-zero compute power. In this paper, we design a scalable distributed inverted list index for disaggregated memory architectures. An inverted list index maps a set of terms to lists of documents that contain this term. Current solutions either partition the index horizontally or vertically with severe limitations in the disaggregated memory setting due to data and access skew, high network latency, or out-of-memory errors. Our method partitions lists into fixed-size blocks and spreads them across the memory nodes to balance skewed accesses. Block-based list processing keeps the memory footprint of compute nodes low and masks latency by interleaving remote accesses with expensive list operations. In addition, we propose efficient updates with optimistic concurrency control and read-write conflict detection. Our experiments confirm the efficiency and scalability of our method.

Asynchronous Privacy-Preservation Federated Learning Method for Mobile Edge Network in Industrial Internet of Things Ecosystem

The typical industrial Internet of Things (IIoT) network system relies on a real-time data upload for timely processing. However, the incidence of device heterogeneity, high network latency, or a malicious central server during transmission has a propensity for privacy leakage or loss of model accuracy. Federated learning comes in handy, as the edge server requires less time and enables local data processing to reduce the delay to the data upload. It allows neighboring edge nodes to share data while maintaining data privacy and confidentiality. However, this can be challenged by a network disruption making edge nodes or sensors go offline or experience an alteration in the learning process, thereby exposing the already transmitted model to a malicious server that eavesdrops on the channel, intercepts the model in transit, and gleans the information, evading the privacy of the model within the network. To mitigate this effect, this paper proposes asynchronous privacy-preservation federated learning for mobile edge networks in the IIoT ecosystem (APPFL-MEN) that incorporates the iteration model design update strategy (IMDUS) scheme, enabling the edge server to share more real-time model updates with online nodes and less data sharing with offline nodes, without exposing the privacy of the data to a malicious node or a hack. In addition, it adopts a double-weight modification strategy during communication between the edge node and the edge server or gateway for an enhanced model training process. Furthermore, it allows a convergence boosting process, resulting in a less error-prone, secured global model. The performance evaluation with numerical results shows good accuracy, efficiency, and lower bandwidth usage by APPFL-MEN while preserving model privacy compared to state-of-the-art methods.

Telemedicine network latency management system in 5G telesurgery: a feasibility and effectiveness study.

Network latency is the most important factor affecting the performance of telemedicine. The aim of the study is to assess the feasibility and efficacy of a novel network latency management system in 5G telesurgery. We conducted 20 telesurgery simulation trials (hitching rings to columns) and 15 remote adrenalectomy procedures in the 5G network environment. Telemedicine Network Latency Management System and the traditional "Ping command" method (gold standard) were used to monitor network latency during preoperative simulated telesurgery and formal telesurgery. We observed the working status of the Telemedicine Network Latency Management System and calculated the difference between the network latency data and packet loss rate detected by the two methods. In addition, due to the lower latency of the 5G network, we tested the alert function of the system using the 4G network with relatively high network latency. The Telemedicine Network Latency Management System showed no instability during telesurgery simulation trials and formal telesurgery. After 20 telesurgery simulation trials and 15 remote adrenalectomy procedures, the p-value for the difference between the network latency data monitored by the Telemedicine Network Latency Management System and the "Ping command" method was greater than 0.05 in each case. Meanwhile, the surgeons reported that the Telemedicine Network Latency Management System had a friendly interface and was easy to operate. Besides, when the network latency exceeded a set threshold, a rapid alarm sounded in the system. The Telemedicine Network Latency Management System was simple and easy to operate, and it was feasible and effective to use it to monitor network latency in telesurgery. The system had an intuitive and concise interface, and its alarm function increased the safety of telesurgery. The system's own multidimensional working ability and information storage capacity will be more suitable for telemedicine work.

Edge computing-based Generative Adversarial Network for photo design style transfer using conditional entropy distance

In recent years, cell phone photography systems have made significant advancements due to their widespread adoption. However, processing the large amount of data generated by these systems on the cloud often leads to high network latency and bandwidth requirements, limiting computational efficiency. To address this challenge, this study introduces a novel design style transfer method for generating high-quality stylized photos in cell phone photography systems using Generative Adversarial Networks (GANs) with an edge computing-based approach. The proposed method aims to enhance the visual appeal of cell phone images by mimicking the artistic style of paintings through abstraction and emphasis on key features. To achieve this objective, a conditional GAN is trained on meticulously labeled data pairs that identify crucial regions in images. This approach offers an efficient solution for computationally intensive tasks by utilizing inference on edge computing nodes. This effectively reduces network latency and bandwidth requirements while improving computational efficiency. Additionally, a penalty is introduced using conditional entropy distance to ensure consistent entropy levels across different photo samples, resulting in higher quality generated photos. Experimental results demonstrate that the proposed CEDGAN model achieves a detection accuracy ranging from 95% to 98%, with consistent results, thus laying a solid foundation for effective photo enhancement. When collecting important portions of materials, CEDGAN maintains a detection accuracy of around 90% to 95% without any significant difference in overall performance. As a result, this method opens up new possibilities for cell phone photography enthusiasts who desire to effortlessly create artistic and stylized photos using edge computing-based GANs. It offers a practical solution for combining technological innovation with creativity in photography.

Secure Decentralized Image Classification With Multiparty Homomorphic Encryption

Decentralized image classification plays a key role in various scenarios due to its attractive properties, including tolerating high network latency and less prone to single-point failures. Unfortunately, training such a decentralized image classification model is more vulnerable to data privacy leaks compared to other distributed training frameworks. Existing efforts exclusively use differential privacy as the cornerstone to alleviate the threat to data privacy. However, differential privacy is implemented at the expense of accuracy, which goes against our motivation for designing an image classification model without loss of accuracy. To address this problem, we propose D <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -MHE, the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">first</i> secure and efficient decentralized training framework with lossless precision. Inspired by the latest developments in the homomorphic encryption technology, we design a multiparty version of Brakerski-Fan-Vercauteren (BFV), one of the most advanced cryptosystems, and use it to implement private gradient updates of users’ local models. D <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -MHE can reduce the communication complexity of general Secure Multiparty Computation (MPC) tasks from quadratic to linear in the number of users, making it very suitable and scalable for large-scale decentralized learning systems. Moreover, D <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -MHE provides strict semantic security protection even if the majority of users are dishonest with collusion. We conduct extensive experiments on MNIST, CIFAR-10, and ImageNet to demonstrate the superiority of D <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -MHE. Experimental results show that D <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -MHE achieves up to 5.5× reduction in computation overhead, and at least 12× reduction in communication overhead compared to existing schemes.

Round-Efficient Secure Inference Based on Masked Secret Sharing for Quantized Neural Network

Existing secure multiparty computation protocol from secret sharing is usually under this assumption of the fast network, which limits the practicality of the scheme on the low bandwidth and high latency network. A proven method is to reduce the communication rounds of the protocol as much as possible or construct a constant-round protocol. In this work, we provide a series of constant-round secure protocols for quantized neural network (QNN) inference. This is given by masked secret sharing (MSS) in the three-party honest-majority setting. Our experiment shows that our protocol is practical and suitable for low-bandwidth and high-latency networks. To the best of our knowledge, this work is the first one where the QNN inference based on masked secret sharing is implemented.

Data Utilization-Based Adaptive Data Management Method for Distributed Storage System in WAN Environment

Recently, research on a distributed storage system that efficiently manages a large amount of data has been actively conducted following data production and demand increase. Physical expansion limits exist for traditional standalone storage systems, such as I/O and file system capacity. However, the existing distributed storage system does not consider where data is consumed and is more focused on data dissemination and optimizing the lookup cost of data location. And this leads to system performance degradation due to low locality occurring in a Wide Area Network (WAN) environment with high network latency. This problem hinders deploying distributed storage systems to multiple data centers over WAN. It lowers the scalability of distributed storage systems to accommodate data storage needs. This paper proposes a method for distributing data in a WAN environment considering network latency and data locality to solve this problem and increase overall system performance. The proposed distributed storage method monitors data utilization and locality to classify data temperature as hot, warm, and cold. With assigned data temperature, the proposed algorithm adaptively selects the appropriate data center and places data accordingly to overcome the excess latency from the WAN environment, leading to overall system performance degradation. This paper also conducts simulations to evaluate the proposed and existing distributed storage methods. The result shows that our proposed method reduced latency by 38% compared to the existing method. Therefore, the proposed method in this paper can be used in large-scale distributed storage systems over a WAN environment to improve latency and performance compared to existing methods, such as consistent hashing.

Prioritized Fault Recovery Strategies for Multi-Access Edge Computing Using Probabilistic Model Checking

The advent of Multi-Access Edge Computing (MEC) has enabled service providers to mitigate high network latencies often encountered in accessing cloud services by deploying containerized application instances on edge servers situated near end users. MEC servers are, however, susceptible to various types of failures such as communication link failures, hardware failures and so on. A fault recovery strategy determines which MEC servers to utilize to re-deploy application containers in the event of a failure. In this work, we propose a two-fold fault recovery strategy characterized by application priority. We propose a Formal Methods driven local recovery strategy for high-priority applications. We use Stochastic Multi-Player Games as a Formal Model to characterize the interactions between the different components in an MEC environment. We use objectives specified in Probabilistic Alternating-Time Temporal Logic with a Probabilistic Model Checker to derive recovery strategies considering all possible execution scenarios of the model. For lower priority applications, we resort to a global recovery strategy by designing a greedy heuristic considering each server’s failure probability. We use benchmark datasets to validate our approach. Experimental results show an average 14% reduction in latency with our approach in comparison with other state-of-the-art methods.

SNAG: SDN-Managed Network Architecture for GridFTP Transfers Using Application-Awareness

Increasingly, academic campus networks support large-scale data transfer workflows for data-intensive science. These data transfers rely on high-performance, scalable, and reliable protocols for moving large amounts of data over a high-bandwidth, high-latency network. GridFTP is a widely used protocol for wide area network (WAN) data movement. However, as the GridFTP protocol does not share connection information with the network-layer, network operators have reduced flexibility, particularly in identifying/managing flows across the network. We address this problem by deploying a production “ <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">application-aware</i> ” software defined network (SDN) for managing GridFTP transfers for data-intensive science workflows. We first propose a novel application-aware architecture called SNAG (SDN-managed Network Architecture for GridFTP transfers). SNAG combines <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">application-layer and network-layer collaboration</i> (termed “application-awareness”) with SDN-enabled network management to classify, monitor and to manage network resources actively. Until now, our SNAG deployment has successfully classified over <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1.5 Billion</i> GridFTP connections at the Holland Computing Center (HCC), University of Nebraska-Lincoln (UNL). Next, we develop an application-aware SDN system to provide differentiated network services for distributed computing workflows. At HCC, we also demonstrate how our system ensures the quality of service (QoS) for high-throughput workflows such as Compact Muon Solenoid (CMS) and Laser Interferometer Gravitational-Wave Observatory (LIGO). Further, we also demonstrate how application-aware SDN can be exploited to create <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">policy-driven</i> approaches to achieve accurate resource accounting for each workflow. We present strategies for implementing differentiated network services and discuss their capacity improvement benefits. Lastly, we provide some guidelines and recommendations for developing application-aware SDN architectures for general-purpose applications.

Dynamic Resource Allocation for Virtual Network Function Placement in Satellite Edge Clouds

Satellite edge computing has become a promising way to provide computing services for Internet of Things (IoT) users in remote areas, which are out of the coverage of terrestrial networks. Nevertheless, it is not suitable for large-scale IoT users due to the resource limitation of satellites. Cloud computing can provide sufficient available resources for IoT users, but it does not meet delay-sensitive services as high network latency. Satellite edge clouds can facilitate flexible service provisioning for numerous IoT users by incorporating the advantages of edge computing and cloud computing. In this paper, we investigate the dynamic resource allocation problem for virtual network function (VNF) placement in satellite edge clouds. The aim is to minimize the network bandwidth cost and the service end-to-end delay jointly. We formulate the VNF placement problem as an integer non-linear programming problem and then propose a distributed VNF placement (D-VNFP) algorithm to address it. The experiments are conducted to evaluate the performance of the proposed D-VNFP algorithm, where Viterbi and Game theory are considered as the baseline algorithms. The results show that the proposed D-VNFP algorithm is effective and efficient for solving the VNF placement problem in satellite edge clouds.

MOTION – A Framework for Mixed-Protocol Multi-Party Computation

We present MOTION, an efficient and generic open-source framework for mixed-protocol secure multi-party computation (MPC) . MOTION is built in a user-friendly, modular, and extensible way, intended to be used as a tool in MPC research and to increase adoption of MPC protocols in practice. Our framework incorporates several important engineering decisions such as full communication serialization, which enables MPC over arbitrary messaging interfaces and removes the need of owning network sockets. MOTION also incorporates several performance optimizations that improve the communication complexity and latency, e.g., \( 2\times \) better online round complexity of precomputed correlated Oblivious Transfer (OT) . We instantiate our framework with protocols for N parties and security against up to \( N-1 \) passive corruptions: the MPC protocols of Goldreich-Micali-Wigderson (GMW) in its arithmetic and Boolean version and OT-based BMR (Ben-Efraim et al., CCS’16), as well as novel and highly efficient conversions between them, including a non-interactive conversion from BMR to arithmetic GMW. MOTION is highly efficient, which we demonstrate in our experiments. Compared to secure evaluation of AES-128 with \( N=3 \) parties in a high-latency network with OT-based BMR, we achieve a 16 \( \times \) better throughput of 16 AES evaluations per second using BMR. With this, we show that BMR is much more competitive than previously assumed. For \( N=3 \) parties and full-threshold protocols in a LAN, MOTION is \( 10\times \) – \( 18\times \) faster than the previous best passively secure implementation from the MP-SPDZ framework, and \( 190\times \) – \( 586\times \) faster than the actively secure SCALE-MAMBA framework. Finally, we show that our framework is highly efficient for privacy-preserving neural network inference.

Immunized Token-Based Approach for Autonomous Deployment of Multiple Mobile Robots in Burnt Area

Collaborative exploration, sensing and communication in previously unknown environments with high network latency, such as outer space, battlefields and disaster hit areas are promising in multi-agent applications. When disasters such as large fires or natural disasters occur, previously established networks might be destroyed or incapacitated. In these cases, multiple autonomous mobile robots (AMR) or autonomous unmanned ground vehicles carrying wireless devices and/or thermal sensors can be deployed to create an end-to-end communication and sensing coverage to support rescue efforts or access the severity of damage. However, a fundamental problem is how to rapidly deploy these mobile agents in such complex and dynamic environments. The uncertainties introduced by the operational environment and wide range of scheduling problem have made solving them as a whole challenging. In this paper, we present an efficient decentralized approach for practical mobile agents deployment in unknown, burnt or disaster hit areas. Specifically, we propose an approach that combines methods from Artificial Immune System (AIS) with special token messages passing for a team of interconnected AMR to decide who, when and how to act during deployment process. A distributed scheme is adopted, where each AMR makes its movement decisions based on its local observation and a special token it receives from its neighbors. Empirical evidence of robustness and effectiveness of the proposed approach is demonstrated through simulation.

A User-Adaptive Layer Selection Framework for Very Deep Sequential Recommender Models

Sequential recommender systems (SRS) have become a research hotspot in recent studies. Because of the requirement in capturing user's dynamic interests, sequential neural network based recommender models often need to be stacked with more hidden layers (e.g., up to 100 layers) compared with standard collaborative filtering methods. However, the high network latency has become the main obstacle when deploying very deep recommender models into a production environment. In this paper, we argue that the typical prediction framework that treats all users equally during the inference phase is inefficient in running time, as well as sub-optimal in accuracy. To resolve such an issue, we present SkipRec, an adaptive inference framework by learning to skip inactive hidden layers on a per-user basis. Specifically, we devise a policy network to automatically determine which layers should be retained and which layers are allowed to be skipped, so as to achieve user-specific decisions. To derive the optimal skipping policy, we propose using gumbel softmax and reinforcement learning to solve the non-differentiable problem during backpropagation. We perform extensive experiments on three real-world recommendation datasets, and demonstrate that SkipRec attains comparable or better accuracy with much less inference time.

Improving latency in Internet-of-Things and cloud computing for real-time data transmission: a systematic literature review (SLR)

To store, analyse and process the large volume of data generated by IoT traditional cloud computing, is used everywhere. However, the traditional cloud data centres have their limitations to handle high latency issues in time-critical applications of IoT and cloud. Their applications are computer gaming, e-healthcare, telemedicine and robot surgery. The high latency in IoTs and cloud includes high computational, communication latency (service) and network latencies. The vital requirement of IoT is to have minimum network, service and computation latencies for real-time applications. Network latency causes a delay in transmitting a message or communication from one location to another. Services that require data in real-time are almost impossible to access the data via the cloud. Traditional cloud computing approaches are unable to fulfil the quality-of-service (QoS) requirements in IoT devices. Researches related to latency reduction techniques are still in infancy. Some new approaches to minimize the latency for transmitting time-sensitive data in real-time are discussed in this paper for cloud and IoT devices. This research will help the researchers and industries to identify the techniques and technologies to minimize the latencies in IoT and cloud. The paper also discusses the research trends and the technical differences between the various technologies and techniques. With the increasing interest in the literature on latency minimization and its requirements for time-sensitive applications; it is important to systematically review and synthesize the approaches, tools, challenges and techniques to minimize latencies in IoT and cloud. This paper aims at systematically reviewing the state of the art of latency minimization to classify approaches, and techniques. The paper uses a PRISMA technique for a systematic review. The paper further identifies challenges and gaps in this regard for future research. We have identified 23 approaches and 32 technologies associated with latencies in the cloud and IoT. A total of 112 papers on latency reduction have been examined under this study. The existing research gaps and works for latency reduction in IoTs are discussed in detail. There are several challenges and gaps, which requires future research work for improving the latency minimization techniques and technologies. Finally, we present some open issues which will determine the future research direction.

Computation Offloading for Smart Devices in Fog-Cloud Queuing System

ABSTRACT Advancements in sensor and hardware technology have surged the growth of smart devices (SDs), including smartphones, and wearable devices. The data generated by the built-in sensors are utilized by different applications such as health-care, smart-city, and connected-vehicles. However, due to the computation and energy limitations of the SDs, they often need to offload the computation-intensive tasks for processing to the remote server. The cloud-based offloading can meet various applications’ demands, but due to high network latency, it is inefficient for real-time applications. Fog computing provides an alternative for the same, as it aggregates the fog nodes’ resources at the edge of the network to meet the computational requirements of the real-time applications. In this paper, we consider a Fog-Cloud architecture consisting of multiple SDs, fog nodes, and the cloud. We use appropriate queuing models to simulate the traffic delay at different network elements and formulate a non-linear multi-objective optimization problem to minimize the energy consumption, execution delay, and cost of remote execution. Finally, the Stochastic Gradient descent (SGD) algorithm based solution approach is proposed that jointly optimizes offloading probability and transmission power to find the optimal trade-off between the offloading objectives. Simulation results show the effectiveness and the efficiency of the proposed system validated by the results.

Exploiting Motion Perception in Depth Estimation Through a Lightweight Convolutional Neural Network

Understanding the surrounding 3D scene is of the utmost importance for many robotic applications. The rapid evolution of machine learning techniques has enabled impressive results when depth is extracted from a single image. High-latency networks are required to achieve these performances, rendering them unusable for time-constrained applications. This article introduces a lightweight Convolutional Neural Network (CNN) for depth estimation, NEON, designed for balancing both accuracy and inference times. Instead of solely focusing on visual features, the proposed methodology exploits the Motion-Parallax effect to combine the apparent motion of pixels with texture. This research demonstrates that motion perception provides crucial insight about the magnitude of movement for each pixel, which also encodes cues about depth since large displacements usually occur when objects are closer to the imaging sensor. NEON's performance is compared to relevant networks in terms of Root Mean Squared Error (RMSE), the percentage of correctly predicted pixels ( δ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> ) and inference times, using the KITTI dataset. Experiments prove that NEON is significantly more efficient than the current top ranked network, estimating predictions 12 times faster; while achieving an average RMSE of 3.118 m and a δ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> of 94.5%. Ablation studies demonstrate the relevance of tailoring the network to use motion perception principles in estimating depth from image sequences, considering that the effectiveness and quality of the estimated depth map is similar to more computational demanding state-of-the-art networks. Therefore, this research proposes a network that can be integrated in robotic applications, where computational resources and processing-times are important constraints, enabling tasks such as obstacle avoidance, object recognition and robotic grasping.

IOT and context‐aware learning‐based optimal neural network model for real‐time health monitoring

AbstractRecently, numerous context‐aware approaches are established to provide physiological information regarding the wellness and healthcare of each individual. While monitoring the health condition of the patient there occur delays in transferring data to the cloud. So to overcome such types of delays, numerous IoT sensors are developed to monitor, track, and sense the activities of the elder persons. This paper proposes four‐module architecture comprises of IoT module (IoT‐M), data pre‐processing module (DP‐M), context‐aware module (CA‐M) as well as decision‐making module (DM‐M) for storing and processing numerous cumulative sensor data. Here, an IoT comprises a concrete or substantial hardware ecological unit, that is, sensors and actuators. On the other hand, the context‐aware computational approach comprises of insubstantial software ecological unit in understanding and processing the context directly into an accomplishment through various IoT devices. The initial module or the IoT‐M comprises of sensors whereas in the DP‐M phase includes data collection phase, data storage phase, and data redundancy phase. The third phase or the CA‐M comprises two different types of layers namely fog layer, cloud layer. In addition to this, a context‐aware learning phase is also enumerated. In the final phase or the DM‐P phase, the feature extraction and classification is done by Back‐Propagation Neural Network along with the Adaptive grasshopper optimization algorithm so as to obtain a best optimal solution. Thus, an alarm or a notification is sent to the medical practitioner regarding the health condition of the patient with very less response time, high accuracy, and a high scalability rate. The evaluation results and discussions are made by comparing our proposed approach with several other approaches and the evaluation results reveal that the proposed framework provides better results with high accuracy, scalability, network latency, and low response time.

Artificial Intelligence for Securing IoT Services in Edge Computing: A Survey

With the explosive growth of data generated by the Internet of Things (IoT) devices, the traditional cloud computing model by transferring all data to the cloud for processing has gradually failed to meet the real-time requirement of IoT services due to high network latency. Edge computing (EC) as a new computing paradigm shifts the data processing from the cloud to the edge nodes (ENs), greatly improving the Quality of Service (QoS) for those IoT applications with low-latency requirements. However, compared to other endpoint devices such as smartphones or computers, distributed ENs are more vulnerable to attacks for restricted computing resources and storage. In the context that security and privacy preservation have become urgent issues for EC, great progress in artificial intelligence (AI) opens many possible windows to address the security challenges. The powerful learning ability of AI enables the system to identify malicious attacks more accurately and efficiently. Meanwhile, to a certain extent, transferring model parameters instead of raw data avoids privacy leakage. In this paper, a comprehensive survey of the contribution of AI to the IoT security in EC is presented. First, the research status and some basic definitions are introduced. Next, the IoT service framework with EC is discussed. The survey of privacy preservation and blockchain for edge-enabled IoT services with AI is then presented. In the end, the open issues and challenges on the application of AI in IoT services based on EC are discussed.

Hierarchical Classification for Constrained IoT Devices: A Case Study on Human Activity Recognition

The massive number of Internet-of-Things (IoT) devices generates a hard-to-manage volume of data. Cloud-centric processing approaches for the IoT data suffer from high and unpredictable network latency, which causes poor experience in real-time IoT applications, such as healthcare. To address this issue, in edge computing, the data inference starts from the data source (i.e., the IoT devices). However, the constrained computational capabilities of the IoT device and the power-hungry data transmission demand a tradeoff between onboard processing and computation offloading. Hence, the IoT information inference requires efficient and lightweight techniques that are tailored for this tradeoff and respect the constrained resources on IoT devices, such as wearables. This article presents a hierarchical classification approach that decomposes the problem into three classifiers in two hierarchy layers. In the first layer, a lightweight classifier executes directly on the IoT device and decides whether to offload the computation to the gateway or to perform it onboard. The second layer comprises a lightweight classifier on the IoT device (can only distinguish a subset of classes) and a complex classifier on the gateway (to distinguish the remaining classes). The experimental results (using a real-world data set for human activity recognition and implemented on a wearable IoT device) show higher accuracy (92% on average) than a nonhierarchical classifier (87% on average). The execution time and power measurements on the IoT device show $3\times $ energy saving for the classification.