Training Rounds Research Articles

Federated deep learning models for detecting RPL attacks on large-scale hybrid IoT networks

With the rapid spread of the Internet of Things (IoT), smart applications and services become increasingly crucial, making them an easily accessible source of personally identifiable information. Over the last few years, the use of machine learning in securing routing layers, particularly routing protocol for low-power and lossy networks (RPL), has become fundamental in ensuring successful routing and privacy preservation as a crucial consideration among edge nodes. In recent works, training of collected data on a central server has increased concerns regarding data privacy. Consequently, decentralized learning is currently a solution for privacy preservation. It has gained popularity in IoT networks in which the models are trained on hybrid data located in edge nodes and enable global decision-making without sharing global data, causing high communication costs during weight updates. We propose a federated learning of routing protocol (Fed-RPL)-based gated recurrent unit (GRU) model for decentralized training rounds and quantization method (Q-8bit) to decrease the number of weight updates that can significantly mitigate the communication overhead and maintain the local model with high accuracy. Meanwhile, the ensemble unit aggregates the updates and selects the best local model to enhance the global model accuracy. Our experiments show that Fed-RPL outperforms classical machine learning (ML) methods in privacy-preserving edge data, significantly reduces the communication cost in non-IID scenarios, and achieves higher detection accuracy than recent FL approaches.

Dynamic differential privacy-based dataset condensation

With the continuous expansion of data scale, data condensation technology has emerged as a means to reduce costs related to storage, time, and energy consumption. Data condensation can generate a synthesized dataset of reduced size, enabling the training of models that exhibit high performance comparable to the original dataset. Nevertheless, data condensation has also exposed privacy issues. Although many approaches have been proposed to preserve privacy for data condensation, the privacy protection for data condensation has not been well explored. Furthermore, to the best of our knowledge, none of the existing approaches propose dynamic parameters-based differential privacy dataset condensation considering unnecessary noise introduced by the fixed privacy parameter strategy. Most approaches typically inject constant noise with the fixed variance into gradients across all layers using predefined privacy parameters, which can significantly impact model accuracy. In this paper, we investigate alternative approaches for data condensation with differential privacy (DP) that aim to ensure DP while minimizing the noise added to gradients and improving the model accuracy. First, we develop a dynamic threshold method to reduce the noise added to gradients in the later stages of training by using a clipping threshold that decreases with training rounds. Second, noise injection in our method is not arbitrary as in conventional approaches; instead, it is based on the maximum size of the gradient after clipping. This approach ensures that only minimal noise increments are introduced, thereby mitigating accuracy loss and parameter instability that may arise from excessive noise injection. Finally, our privacy analysis confirms that the proposed method provides a rigorous privacy guarantee. Extensive evaluations on different datasets demonstrate that our approach can improve accuracy compared to existing DP data condensation techniques while adhering to the same privacy budget and applying a specified clipping threshold.

The analysis of educational informatization management learning model under the internet of things and artificial intelligence

This study explores the influence of the Internet of Things (IoT) and Artificial Intelligence (AI)-enhanced learning models on student management in educational informatization management. A game-theoretic enhanced learning model is proposed to achieve this objective, incorporating resource scheduling strategies under fog computing and a student management system that integrates IoT and AI technologies. This model’s performance and the student management system are then tested. The results indicate that the fog computing-based hierarchical Q-learning (Q) model proposed in this study achieves faster convergence than a single Q model, reaching convergence after 80 training rounds, ten rounds earlier than the comparative algorithm. The model exhibits a lower average workload delay of 0.5 ms and fog node delay below 1 ms, showcasing significant advantages in terms of overall cost-effectiveness, thus minimizing service costs. The student management system has 3000 concurrent user connections, static page request times ranging from 0 to 25 s, login response time predominantly at 60 s, and a capacity to process up to 20 parallel tasks per second with zero errors. The system functionalities are fully realized, meeting usage demands effectively and achieving the highest average functional score of 9.03 for online interaction functionality. This study demonstrates the efficacy of the game-theoretic enhanced learning model in a fog computing environment and the positive impact of IoT and AI technologies on student management. The proposed student management system better caters to individual student needs, enhancing learning outcomes and experiences. The study's innovation lies in the integration of IoT technology with AI-enhanced learning models, coupled with the introduction of game-theoretic resource scheduling strategies, enabling the student management system to intelligently identify student requirements, allocate learning resources, and dynamically optimize the educational process, ultimately improving learning outcomes. This holds significant implications for enhancing education quality and promoting personalized student development.

Digital Twin and federated learning enabled cyberthreat detection system for IoT networks

The widespread deployment of Internet of Things (IoT) devices across various smart city applications presents significant security challenges, increased by the rapidly evolving landscape of cyber threats. Traditional security solutions, including those using Federated Learning with federated averaging, suffer from inefficiencies due to random node selection and partial data sampling, which can hinder the detection of comprehensive network-wide attacks. This paper introduces a novel Cyberthreat Detection System for IoT networks that leverages Digital Twin technology and an optimized Federated Learning approach. Our hypothesis implies integrating Digital Twin models within an IoT security framework to improve real-time cyberthreat detection capabilities. We implement a 'Adaptive Thresholding with Early Stopping method' based methodology in Federated Learning to systematically train and aggregate local models based on predefined training rounds, thereby ensuring that all local models contribute to the global model until a target accuracy is achieved. This method significantly improves the detection of zero-day attacks by reducing dependency on random selections and partial data. The system architecture features Digital Twins of IoT medical infrastructure components—such as radiology, intensive care, and outpatient care—positioned at the network edge to minimize latency and bandwidth usage. Comparative evaluations of our model against traditional federated averaging methods demonstrate superior performance, with enhancements in model aggregation efficiency evidenced by higher F1 scores and reduced CPU usage. Specifically, our distributed digital twin environment at the edge layer shows 14% and 33% latency reductions compared to fog and cloud-based implementations, respectively. This study highlights the potential of Digital Twin and advanced Federated Learning methodologies to secure IoT networks against evolving and growing cyber threats.

Eco-FL: Enhancing Federated Learning sustainability in edge computing through energy-efficient client selection

In the realm of edge cloud computing (ECC), Federated Learning (FL) revolutionizes the decentralization of machine learning (ML) models by enabling their training across multiple devices. In this way, FL preserves privacy and minimizes the need for centralized data by processing data near the source. From a communication standpoint, only the model weights are exchanged between devices. By avoiding the need to send data to a centralized location for processing, FL reduces the energy required for data transfer and supports more efficient use of computing resources at the edge. FL is particularly advantageous for resource-constrained devices, such as smartphones and IoT devices. However, this limited computational power and battery capacity and the challenge of energy consumption are critical aspects of FL systems. This paper introduces Eco-FL, an innovative methodology designed to optimize energy consumption in FL systems, in the field of Green Edge Cloud Computing (GECC). Our approach employs a device selection process that considers the entropy of the data held by the devices and their available energy reserves. This ensures that devices with lower energy availability are less likely to participate in the training rounds, prioritizing those with higher energy capacities. To evaluate the efficacy of our methodology, we utilize FedEntropy, an entropy-based aggregation method, alongside established aggregation methods such as FedAvg and FedProx for performance comparison. The effectiveness of Eco-FL in reducing energy consumption without compromising the accuracy of the FL process is demonstrated through analyses conducted on three distinct datasets. These analyses vary the β parameter of the Dirichlet distribution and account for scenarios with both homogeneous and heterogeneous initial device charges. Our findings validate Eco-FL’s potential to enhance the sustainability of FL systems by judiciously managing client participation based on energy criteria, presenting a significant step forward in the development of energy-efficient FL.

Client selection and resource scheduling in reliable federated learning for UAV-assisted vehicular networks

Federated Learning (FL), a promising deep learning paradigm extensively deployed in Vehicular Edge Computing Networks (VECN), allows a distributed approach to train datasets of nodes locally, e.g., for mobile vehicles, and exchanges model parameters to obtain an accurate model without raw data transmission. However, the existence of malicious vehicular nodes as well as the inherent heterogeneity of the vehicles hinders the attainment of accurate models. Moreover, the local model training and model parameter transmission during FL exert a notable energy burden on vehicles constrained in resources. In view of this, we investigate FL client selection and resource management problems in FL-enabled UAV-assisted Vehicular Networks (FLVN). We first devise a novel reputation-based client selection mechanism by integrating both data quality and computation capability metrics to enlist reliable high-performance vehicles. Further, to fortify the FL reliability, we adopt the consortium blockchain to oversee the reputation information, which boasts tamper-proof and interference-resistant qualities. Finally, we formulate the resource scheduling problem by jointly optimizing the computation capability, the transmission power, and the number of local training rounds, aiming to minimize the cost of clients while guaranteeing accuracy. To this end, we propose a reinforcement learning algorithm employing an asynchronous parallel network structure to achieve an optimized scheduling strategy. Simulation results show that our proposed client selection mechanism and scheduling algorithm can realize reliable FL with an accuracy of 0.96 and consistently outperform the baselines in terms of delay and energy consumption.

Crack recognition and defect detection of assembly building constructions for intelligent construction

Vision-assisted surface defect detection technology is shallowly applied in crack identification of assembly building components, for this reason, the study proposes a crack identification and defect detection method for assembly building components oriented to intelligent construction. An image preprocessing algorithm is designed by improving bilateral filtering, on the basis of which an image classification model is constructed using the GhostNet algorithm, and the cracks are localized and measured using the 2D pixel positioning technique. Algorithm validation showed that the processed image denoising is better, and the peak signal-to-noise ratio of the image of the proposed algorithm is improved by 15.701 % and 2.395 %, respectively, compared to other algorithms. The F1 value of the proposed model after 50 training rounds increased by 20.970 % on average compared to other models, and the detection accuracy was as high as 0.990. The actual measurements of cracks in concrete wall panels revealed that the research-proposed method has better results compared to the traditional manual measurements, and is not subject to the limitations and interferences of factors such as manual experience, and it is more effective in the recognition of crack images. Overall, the detection method proposed by the study has high accuracy and small error, can meet the needs and standards of crack detection in assembly building components, and can intelligently locate the maximum length and width coordinates of the cracks, which is of high value in the application of crack detection in assembly building components.

Classical music recommendation algorithm on art market audience expansion under deep learning

Abstract The purpose of the study is to help users know about their favorite music and expand art market audiences. First, the personalized recommendation data of classical music are obtained based on the deep learning recommendation algorithm technology, artificial intelligence, and music playback software of users. Second, a systematic experiment is conducted on the improved recommendation algorithm, and a classical music dataset is established and used for model training and user testing. Then, the network model of the classical music recommendation algorithm is constructed through the typical convolutional neural network model, and the optimal parameters suitable for the model are found. The experimental results show that the optimal value of the dimension in the hidden layer is 192, and 24,000 training rounds can converge to the global optimum when the learning rate is 0.001. The personalized recommendation is provided for target users by calculating the similarity between user preference and potential features of classical music, relieving the auditory fatigue of art market audiences, improving user experience, and expanding the art market audience through the classical music recommendation system.

Equilibrium in the Computing Continuum through Active Inference

Computing Continuum (CC) systems are challenged to ensure the intricate requirements of each computational tier. Given the system’s scale, the Service Level Objectives (SLOs), which are expressed as these requirements, must be disaggregated into smaller parts that can be decentralized. We present our framework for collaborative edge intelligence, enabling individual edge devices to (1) develop a causal understanding of how to enforce their SLOs and (2) transfer knowledge to speed up the onboarding of heterogeneous devices. Through collaboration, they (3) increase the scope of SLO fulfillment. We implemented the framework and evaluated a use case in which a CC system is responsible for ensuring Quality of Service (QoS) and Quality of Experience (QoE) during video streaming. Our results showed that edge devices required only ten training rounds to ensure four SLOs; furthermore, the underlying causal structures were also rationally explainable. The addition of new types of devices can be done a posteriori; the framework allowed them to reuse existing models, even though the device type had been unknown. Finally, rebalancing the load within a device cluster allowed individual edge devices to recover their SLO compliance after a network failure from 22% to 89%.

MADDPG-Based Offloading Strategy for Timing-Dependent Tasks in Edge Computing

With the increasing popularity of the Internet of Things (IoT), the proliferation of computation-intensive and timing-dependent applications has brought serious load pressure on terrestrial networks. In order to solve the problem of computing resource conflict and long response delay caused by concurrent application service applications from multiple users, this paper proposes an improved edge computing timing-dependent, task-offloading scheme based on Multi-Agent Deep Deterministic Policy Gradient (MADDPG) that aims to shorten the offloading delay and improve the resource utilization rate by means of resource prediction and collaboration among multiple agents to shorten the offloading delay and improve the resource utilization. First, to coordinate the global computing resource, the gated recurrent unit is utilized, which predicts the next computing resource requirements of the timing-dependent tasks according to historical information. Second, the predicted information, the historical offloading decisions and the current state are used as inputs, and the training process of the reinforcement learning algorithm is improved to propose a task-offloading algorithm based on MADDPG. The simulation results show that the algorithm reduces the response latency by 6.7% and improves the resource utilization by 30.6% compared with the suboptimal benchmark algorithm, and it reduces nearly 500 training rounds during the learning process, which effectively improves the timeliness of the offloading strategy.

Enhancing Semi-Supervised Few-Shot Hyperspectral Image Classification via Progressive Sample Selection

Hyperspectral images (HSIs) provide valuable spatial–spectral information for ground analysis. However, in few-shot (FS) scenarios, the limited availability of training samples poses significant challenges in capturing the sample distribution under diverse environmental conditions. Semi-supervised learning has shown promise in exploring the distribution of unlabeled samples through pseudo-labels. Nonetheless, FS HSI classification encounters the issue of high intra-class spectral variability and inter-class spectral similarity, which often lead to the diffusion of unreliable pseudo-labels during the iterative process. In this paper, we propose a simple yet effective progressive pseudo-label selection strategy that leverages the spatial–spectral consistency of HSI pixel samples. By leveraging spatially aligned ground materials as connected regions with the same semantic and similar spectrum, pseudo-labeled samples were selected based on round-wise confidence scores. Samples within both spatially and semantically connected regions of FS samples were assigned pseudo-labels and joined subsequent training rounds. Moreover, considering the spatial positions of FS samples that may appear in diverse patterns, to fully utilize unlabeled samples that fall outside the neighborhood of FS samples but still belong to certain connected regions, we designed a matching active learning approach for expert annotation based on the temporal confidence difference. We identified samples with the highest training value in specific regions, utilizing the consistency between predictive labels and expert labels to decide whether to include the region or the sample itself in the subsequent semi-supervised iteration. Experiments on both classic and more recent HSI datasets demonstrated that the proposed base model achieved SOTA performance even with extremely rare labeled samples. Moreover, the extended version with active learning further enhances performance by involving limited additional annotation.

Statistical assessment of reliability of anthropometric measurements in the multi-site South African National Dietary Intake Survey 2022.

Anthropometric data quality in large multicentre nutrition surveys is seldom adequately assessed. In preparation for the South African National Dietary Intake Survey (NDIS-2022), this study assessed site leads' and fieldworkers' intra- and inter-rater reliability for measuring weight, length/height, mid-upper arm circumference (MUAC), waist circumference (WC) and calf circumference (CC). Standardised training materials and measurement protocols were developed, and new anthropometric equipment was procured. Following two training rounds (12 site lead teams, 46 fieldworker teams), measurement reliability was assessed for both groups, using repeated measurements of volunteers similar to the survey target population. Reliability was statistically assessed using the technical error of measurement (TEM), relative TEM (%TEM), intra-class correlation coefficient (ICC) and coefficient of reliability (R). Agreement was visualised with Bland-Altman analysis. By %TEM, the best reliability was achieved for weight (%TEM = 0.260-0.923) and length/height (%TEM = 0.434-0.855), and the poorest for MUAC by fieldworkers (%TEM = 2.592-3.199) and WC (%TEM = 2.353-2.945). Whole-sample ICC and R were excellent ( > 0.90) for all parameters except site leads' CC inter-rater reliability (ICC = 0.896, R = 0.889) and fieldworkers' inter-rater reliability for MUAC in children under two (ICC = 0.851, R = 0.881). Bland-Altman analysis revealed no significant bias except in fieldworkers' intra-rater reliability of length/height measurement in adolescents/adults ( + 0.220 (0.042, 0.400) cm). Reliability was higher for site leads vs. fieldworkers, for intra-rater vs. inter-rater assessment, and for weight and length/height vs. circumference measurements. NDIS-2022 site leads and fieldworkers displayed acceptable reliability in performing anthropometric measurements, highlighting the importance of intensive training and standardised measurement protocols. Ongoing reliability assessment during data collection is recommended.

Privacy-preserving and scalable federated blockchain scheme for healthcare 4.0

IoT plays a significant role in the growth of clinical data for identifying hazardous diseases and building drugs for patient diagnosis and medical care. Blockchain and federated learning are widely proposed in existing studies for localized data training and storage in a secure, decentralized environment. However, several studies rely on the federated averaging method in federated learning, which introduces high energy consumption during several training rounds. Frequent transactions for storing patient records in blockchain and smart contract processing present network congestion challenges. In this paper, we propose a privacy-preserving federated learning and scalable blockchain scheme. First, we present a satisfaction scoring method for model aggregation to improve energy efficiency during the federated learning process. Secondly, we design an Ethereum-based blockchain network with sidechains to process smart contract transactions separately and reduce computation overload in the blockchain mainchain. Evaluation of the proposed scheme is compared with the baseline federated average method and transaction processing of smart contracts with the Ethereum main chain. Results demonstrate reduced CPU consumption using the proposed model aggregation method, an improvement over the federated average method, and an improvement of 43.65 % in transaction processing speed using two sidechains ensuring side.

FedMon: A Federated Learning Monitoring Toolkit

Federated learning (FL) is rapidly shaping into a key enabler for large-scale Artificial Intelligence (AI) where models are trained in a distributed fashion by several clients without sharing local and possibly sensitive data. For edge computing, sharing the computational load across multiple clients is ideal, especially when the underlying IoT and edge nodes encompass limited resource capacity. Despite its wide applicability, monitoring FL deployments comes with significant challenges. AI practitioners are required to invest a vast amount of time (and labor) in manually configuring state-of-the-art monitoring tools. This entails addressing the unique characteristics of the FL training process, including the extraction of FL-specific and system-level metrics, aligning metrics to training rounds, pinpointing performance inefficiencies, and comparing current to previous deployments. This work introduces FedMon, a toolkit designed to ease the burden of monitoring FL deployments by seamlessly integrating the probing interface with the FL deployment, automating the metric extraction, providing a rich set of system, dataset, model, and experiment-level metrics, and providing the analytic means to assess trade-offs and compare different model and training configurations.

An efficient bandwidth-adaptive gradient compression algorithm for distributed training of deep neural networks

In distributed deep learning with data parallelism, communication bottleneck throttles the efficiency of model training. Recent studies adopt versatile gradient compression techniques, with communication sparsification standing out as an effective approach for reducing the number of gradients to be transmitted. However, the deployment of gradient sparsification is adversely influenced by the change of network environment in real systems, and existing methods either neglect bandwidth dynamics during training or experience drastic fluctuation of compression ratios. In this paper, we propose ACE, a novel adaptive gradient compression mechanism with high communication efficiency under bandwidth variation. ACE adapts the sparsification ratio to the average bandwidth in a time window, other than following its dynamics exactly. To accurately compute the compression ratio, we first profile the compression time and model a single iteration time consisting of communication, computation and compression operations. We then present a practical model to fit the needed training rounds till convergence, and formulate an optimization problem to compute the optimal sparsification ratio. We conduct experiments on different DNN models in different network environments and compare various methods in terms of convergence and model quality. The experimental results show that ACE achieves up to 9.39× and 1.28× training speedups over fixed and state-of-the-art adaptive compression methods.

Research on Small Target Detection Method for Industrial Safety Helmets Based on Improved YOLOv8

Industrial safety helmets are crucial personal protective gear but detecting them as small targets in complex environments is challenging. This work proposes enhancements to the YOLOv8 object detection framework, specifically incorporating a spatial-to-depth (SPD) convolution module and a large selective kernel network (LSKNet). SPD-Conv combines spatial-to- depth layers and non-strided convolutions to retain fine-grained information when downscaling feature maps, while LSKNet introduces dynamic spatial selection and attention for refined context modeling. Our customized model is trained on a dataset of construction hardhat images captured via drones. Quantitative results showcase higher precision and recall over baseline YOLOv8, surpassing competing YOLOv5 versions. An optimized final model outcomes demonstrate accuracy exceeding 90% validation in mAP metric after 200 training rounds. By tackling limitations posed by small, obscured industrial safety gears, this enhanced real-time detection approach provides indispensable technological support for bolstering workplace hazard identification and prevention.

EneA-FL: Energy-aware orchestration for serverless federated learning

Federated Learning (FL) represents the de-facto standard paradigm for enabling distributed learning over multiple clients in real-world scenarios. Despite the great strides reached in terms of accuracy and privacy awareness, the real adoption of FL in real-world scenarios, in particular in industrial deployment environments, is still an open thread. This is mainly due to privacy constraints and to the additional complexity stemming from the set of hyperparameters to tune when employing AI techniques on bandwidth-, computing-, and energy-constrained nodes. Motivated by these issues, we focus on scenarios where participating clients are characterised by highly heterogeneous computing capabilities and energy budgets proposing EneA-FL, an innovative scheme for serverless smart energy management. This novel approach dynamically adapts to optimise the training process while fostering seamless interaction between Internet of Things (IoT) devices and edge nodes. In particular, the proposed middleware provides a containerised software module that efficiently manages the interaction of each worker node with the central aggregator. By monitoring local energy budget, computational capabilities, and target accuracy, EneA-FL intelligently takes informed decisions about the inclusion of specific nodes in the subsequent training rounds, effectively balancing the tripartite trade-off between energy consumption, training time, and final accuracy. Finally, in a series of extensive experiments across diverse scenarios, our solution demonstrates impressive results, achieving between 30% and 60% lower energy consumption against popular client selection approaches available in the literature while being up to 3.5 times more efficient than standard FL solutions.

FedNow: An Efficiency-Aware Incentive Mechanism Enables Privacy Protection and Efficient Resource Utilization in Federated Edge Learning

Federated edge learning (FEL) has recently attracted great interest due to its real-time response and energy-efficient characteristics. Most existing work focuses on designing algorithms to improve model performance, ignoring the malicious behavior and personal decision-making of self-interested edge servers. Although some efforts have been devoted to incentivizing honest edge server engagement by compensating training costs, this rarely considers resource efficiency and often assumes that edge servers provide complete information to the platform, which may lead to the risk of private attribute leakage. Hence, we aim to achieve an incentive mechanism that promotes secure and efficient model training between the platform and edge servers. However, edge servers’ multi-dimensional private attributes and training strategies make the optimization problem nonconvex, and incomplete information further increases the complexity of the analysis. In order to address these challenges and by integrating contract theory and exponential mechanism, we propose an efficiency-aware incentive mechanism, FedNow, which enables edge servers to personally determine their local training rounds while motivating their participation without giving access to their true training strategies and private attributes. Specifically, we enabld edge servers to add noise to their submitted training strategy to hide their true training rounds; then, we carefully designed an efficiency score function to select honest and efficient edge servers without disclosing their private attributes. In order to demonstrate that FedNow strictly outperforms existing schemes in terms of total costs, we theoretically derived sufficient conditions for making the total costs of FedNow lower than existing schemes and designed a greedy algorithm that uses the Monte Carlo method to find feasible near-optimal solutions in polynomial time. Our extensive experimental assessment using synthetic and real datasets shows the superiority of FedNow.

A brain tumour classification on the magnetic resonance images using convolutional neural network based privacy‐preserving federated learning

AbstractThe healthcare industry has found it challenging to build a powerful global classification model due to the scarcity and diversity of medical data. The leading cause is privacy, which restricts data sharing among healthcare providers. Federated learning (FL) can contribute to developing classification models by protecting data privacy. This study has tested various federated techniques in a peer‐to‐peer setting to classify brain Magnetic Resonance Images (MRI). The authors propose various aggregation strategies for FL, including Federated Averaging (FedAvg), Quantum FL with FedAVG (QFedAvg) and Fault Tolerant FedAvg (Ft‐FedAvg) and FedAvg with differential privacy (Dp‐FedAvg). In each approach, a custom Convolutional Neural Network (CNN) model is applied to compute locally run nodes with different parts of the same brain MRI dataset for 10, 20 and 30 training and test rounds. A central server and CNN‐based three federated clients are included in the FL‐based brain tumour classification model to exchange data and combine the model weights on the server, which are sent from local devices to the server. The superiority of the performance of the proposed model is demonstrated by comparing it with traditional methods on various performance metrics. Experimental results show that in brain MRI dataset classification using FL approaches, FedAVg showed the best performance with 85.55% and 84.60% success for 10 and 20 rounds, respectively, while Ft‐FedAvg showed the best performance with 85.80% success for 30 rounds for test set. Statistical results obtained from FL approaches showed that FedAvg and Ft‐FedAvg have superior performance with regard to accuracy and robustness in comparison with the others.

Heterogeneity-aware device selection for efficient federated edge learning

Federated learning (FL) combined with mobile edge computing (FEEL) provides an end-to-edge synergetic learning approach to allow end devices to participate in machine learning model training parallelly while ensuring user privacy is maintained. However, conventional FL approaches often overlook two critical characteristics in the real-edge scenario: system heterogeneity and statistical heterogeneity. This oversight can detrimentally impact both the training efficiency and the model's accuracy. Specifically, it brings intolerable training delays and severe training accuracy degradation. To address these issues, this paper proposes a novel Quality-aware online Device Selection (QDS) algorithm. The QDS algorithm leverages a greedy selection method that guarantees the deadline restrictions and reflects upon their historical performance metrics, as indicated by their loss function values from preceding training rounds. This rigorous selection process ensures that the participating device set is optimally positioned to balance the dual objectives of training efficiency and model accuracy. Furthermore, we have developed a training loss-based device selection mechanism, aimed at prioritizing higher-quality devices for early submission of local updates prior to the designated deadline. Experimental findings demonstrate that the proposed QDS significantly enhances both the speed and accuracy of training when contrasted with baseline methods.