Distributed computing engines, such as Spark, are widely used to process the large volumes of data collected by the Internet of Things (IoT) devices. In IoT systems, computing nodes often exhibit significant heterogeneity. However, most existing task scheduling algorithms neglect the performance differences among system nodes and statically assign tasks based on data locality. When high-performance nodes complete local tasks rapidly, they are subsequently assigned massive non-local tasks, resulting in severe network congestion and performance degradation. To address these issues, we propose a dynamic resource utilization-aware task scheduling strategy (DRUTS), which can efficiently utilize idle bandwidth and high-performance computing resources while also ensuring data locality. Firstly, considering the time-varying characteristics of heterogeneous node performance under various workloads, we use a sliding window to process the task information stream and evaluate the relative performance of nodes in real-time. Then, our proposed strategy adopts weighted random methods to dynamically select taskprovidersandreceiversby combining the cluster network load. Finally, it migrates and pre-executes tasks to high-performance nodes based on data distribution. We evaluate our proposed strategy performance using six typical workloads on two types of real-world heterogeneous clusters. The experimental results clearly demonstrate that our proposed strategy not only improves CPU and network utilization by 15.7% and 13.2%, respectively, but also reduces application execution time by 37.2% compared to existing work.

In the Internet of Vehicles (IoV) systems, recognizing driver emotions is crucial to alleviate dangerous driving behaviors caused by emotional instability. Current research predominantly utilizes multimodal data generated by various types of sensors in IoV systems as input to analyze driver emotion changes using multimodal models. However, existing methods are not enough to fully exploit the advantages of large language models (LLM) in information extraction and multimodal feature fusion, which limits the inference capability of emotion recognition models. Therefore, this article proposes an LLM-auxiliary supervision module, which assists in the training phase through LLM to enhance the performance of multimodal emotion recognition models. Specifically, we designed a label text feature extraction (LTFE) module that employs LLM for text data augmentation and extraction, converting label text into semantically informative feature representations. Additionally, we proposed the label-auxiliary supervision (LAS) strategy, which effectively integrates the LLM label text features learned from the LTFE module with the multimodal emotion recognition model during the training phase to enhance the model’s inference ability. Notably, the LTFE and LAS modules are used only during the training phase, ensuring that the backbone model requires minimal computational resources during inference, making it compatible with the computational constraints of intelligent vehicular devices. Extensive experiments conducted on the PPB-Emo, RAVDESS, and IEMOCAP datasets demonstrate that the proposed method outperforms existing approaches in driver emotion recognition tasks.

Function as a Service (FaaS) is a popular cloud computing service model that incorporates an auto-scaling mechanism, enabling applications to dynamically adjust computing resources, achieving rapid response to load changes and efficient resource utilization. However, the limited resource allocation mode for function containers can frequently cause function performance degradation before scaling is complete, so some FaaS platforms address this issue by default through a shared-resource mode. But existing constant target load-based autoscalers fail to perceive node-level load under this mode, leading to numerous scaling decisions to nodes that have already reached their load bottlenecks, without bringing actual resource or performance gains. This makes the system underutilized and even degrades its performance. To solve this issue, in this paper, we design a horizontal autoscaler, NDScaler, which efficiently scales functions in shared-resource mode by using the node load-aware scaling strategy, thereby eliminating invalid scaling behaviours and the resulting degradation of function performance. We implement this strategy through the proposed node load-aware and dynamic target load algorithm, which models the scale-up problem as a load transfer problem between nodes and functions and adopts a greedy search strategy to identify the optimal target functions for scale-up. Furthermore, it introduces a dynamic load target to assess the extent of load reduction for functions and accurately scales functions down. We have implemented NDScaler and evaluated it in detail on the OpenFaaS platform. Experimental results show that, compared with existing methods, NDScaler can ensure scaling effectiveness and achieve high-efficiency scaling in both simple single-function scenarios and complex multifunction scenarios, effectively improving function throughput while significantly reducing latency.

Federated Learning (FL) has gained widespread attention due to its strong privacy protections and collaborative learning capabilities. Recently, Personalized Federated Learning (PFL) has garnered significant attention for its ability to address statistical heterogeneity. Most existing PFL methods either focus on feature extraction, struggling to balance collaborative learning and personalization, or emphasize dynamic weight adjustments, relying on heuristic designs that lead to lower communication efficiency in large-scale federated learning systems. However, these methods fail to effectively integrate these two aspects to achieve both efficient collaborative learning and personalized goals. To address these issues, this paper proposes an Adaptive Feature-Driven Weighting Based Personalized Federated Learning (FedAFW) approach. FedAFW first utilizes local feature representations to guide the generation of global and personalized weights, enhancing the personalization effect. Subsequently, it uses gradient similarity for weight allocation, balancing the relative contributions of the global and personalized models, thus improving overall performance. Experiments on diverse datasets under heterogeneous settings show that FedAFW improves accuracy by up to 5.84%, boosts communication efficiency by 54.6%, and outperforms advanced methods in scalability and stability, demonstrating its robustness in handling statistical heterogeneity.

Medical image segmentation is crucial for separating tissue structures and anatomical regions. However, due to significant variations in size, shape, and density of target tissues in medical images, this task faces many challenges. Neural networks are widely used in medical image segmentation due to their powerful feature extraction and pattern recognition capabilities. But traditional Convolutional Neural Networks (CNNs) struggle to capture long-range dependencies, and Transformer models may lack sufficient channel interaction and detail representation. To address the above issues, this paper proposes a novel architecture called TSNet, which innovatively integrates SimAM (Neural Attention Module) and Triplet Attention mechanism. First, Triplet Attention adopts a three-branch structure to effectively encodes channel and spatial information. By reducing information loss and achieving direct correspondence between channels and weights, it significantly enhances the model’s feature extraction and representation capabilities in complex medical image processing. Meanwhile, the parameter-free SimAM module generates adaptive 3D attention weights by optimizing the energy function, further optimizing the interaction and fusion between features. Finally, extensive experiments on real datasets for heart and CT segmentation have shown that the proposed TSNet performs significantly better than the baseline method in terms of Dice Similarity Coefficient (DSC) and the 95th percentile Hausdorff Distance (HD95).

With the support for data-intensive, rate-hungry and delay-sensitive applications, 6G enabled massive IoT is surely becoming the most potential computing paradigm. Along with this trend, the scale of mobile devices and data traffic in the network is increasing explosively, resulting in huge transmission pressure on the backbone network, accompanied by serious security problems. All above call for a secure and high-throughput data communication system for 6G enabled massive IoT. In this paper, an Active-Trust based security Service Orchestration (ATSO) framework is proposed. First, the active-trust evaluation mechanism is introduced at the data acquisition layer, and direct trust is combined with indirect trust to accurately evaluate the trust of data providers. Then, service orchestration mechanism is proposed, which orchestrates data into services through edge devices to implement the service-oriented architecture, and conducts progressive aggregation at routing layer to form more advanced services. Extensive simulation results demonstrate that ATSO effectively improve performance in data security, energy efficiency and delay. Finally, we discuss the potential challenges in promoting the study of ATSO.

The proliferation of various IoT devices has brought about diverse computing requests. Scheduling delay-sensitive tasks to edge nodes closer to data sources can help alleviate core network congestion and improve system quality of service (QoS). However, with the dynamic computing requirements of changing scenarios and the imbalanced performance of limited heterogeneous edge resources, resource competition among multiple tasks has become increasingly fierce. This resource competition leads to inefficient services and performance fluctuations in edge scheduling systems. The key lies in dynamically matching task requirements and limited heterogeneous resources to improve resource utilization efficiency. To overcome this challenge, we propose a resource-aware task grouping scheduling strategy (RATGS) based on our proposed group-based and sharedstate edge scheduling framework, aiming to improve the overall service quality of edge computing systems. We perform extensive evaluation on multiple metrics using realistic workloads and realworld traces. The experimental results demonstrate that RATGS improves the task completion rate by 7.56%∼50.1% before the deadline and improves the efficiency of resource utilization by 17.7%∼94.8% compared with existing baseline strategies. In addition, RATGS performed second best in terms of average completion time.

Predicting dust pollution is necessary to achieve good air quality and patient recovery. In the era of big data, there are already many machine learning algorithms for predicting the concentration of air pollutant PM2.5. However, these model methods perform poorly when dealing with massive air quality data sets and do not solve the impact of data distribution imbalance. Therefore, a multi-strategy collaborative feature selection method combining random forests and recursive feature elimination with adaptive boosting rotation forest (RFR-ABROF) algorithm is proposed. On this basis, multiclass adaptive synthetic sampling (Multi-ADASYN) strategy is introduced to balance the imbalanced air quality data set. We verified the effectiveness of the model through comparative experiments, which show our proposed model has the higher values of accuracy, precision, recall, f1-score, and ROC curve area under evaluation indicators in the air quality data set of four locations in most cases, and the more considerable the amount of data, the better the model prediction performance.

The concept drift detection algorithm can timely respond to and adjust the model by monitoring changes in data distribution over time. However, the dynamically adjusted ensemble model may still retain some components with weak adaptability. These components are involved in subsequent training and testing phases, leads to a significant decrease in classification performance. To solve these problems, this paper proposes an Adaptive Hoeffding Tree Model Based on Differential Entropy and Relative Entropy (AHT-DERE) for concept drift detection. It adopts a two-step strategy: a) A differential entropy-based drift detection method, which calculates the information entropy of the two most recently arrived data samples, and quantifies the difference between data distributions by subtracting the entropy values. This measurement serves as the criteria for determining the occurrence of concept drift. b) A relative entropy-based dynamic adjustment method, which utilizes the relative entropy similarity between the fitted and true distributions of the current data samples. This method selects well-adapted components for each round of incremental updates to improve the resilience of the ensemble model to concept drift. Compared to advanced algorithms, experimental results show that in two sets of experiments, the classification performance of AHT-DERE achieved an average improvement of 6.36% and 5.94% on seven publicly available real-world and synthetic datasets, respectively. The maximum improvement reached 13.55% and 10.82%, respectively.

Face template protection has received widespread attention in the field of biometric security. However, most of the recent schemes can not resistant to randomness source exposure, resulting in compromised protected templates, and the verification accuracy needs to be further improved to meet the needs of realistic applications. In this paper, we propose a novel modular architecture called SecureVeil for protecting face templates. SecureVeil protects face templates with a deep cosine transformation network called FlexNet, which performs random orthogonal transforms on face templates using user-specific keys. To protect user-specific keys, SecureVeil employs a secure key fusion construction called SecureFusion, which fuses user-specific keys with face templates and permutation vectors. We evaluate the irreversibility, unlinkability and verification accuracy of SecureVeil on two state-of-the-art face recognition systems, including ArcFace and FaceNet, using three benchmarking datasets, including MOBIO, LFW, and CFP. Experimental results show that the irreversibility of SecureVeil outperforms existing related schemes. Its verification accuracy is superior than all these compared schemes with an average improvement of 7.70%, and improves by 12.45% on FaceNet when using the LFW dataset. Overall, SecureVeil meets the four criteria for face template protection.

With the advantages of maneuverability and low cost, Unmanned Aerial Vehicles (UAVs) are widely deployed in mobile edge computing as micro servers to provide computing service. However, tasks usually require a large amount of energy and have strict time constraints, while the battery energy and endurance of UAVs are limited. Therefore, energy consumption and delay have become key issues in such architectures. To address this issue, an Entropy Normalized Soft Actor-Critic (ENSAC) computation offloading algorithm is proposed in this paper, aiming to minimize the weighted sum of task offloading delay and energy consumption. In ENSAC, we formulate the task offloading problem as a Markov Decision Process (MDP). Considering the non-convexity, high-dimensional state space, and continuous action space of this problem, the ENSAC algorithm fully combines deviation strategy and maximum entropy reinforcement learning, and designs a system utility function under entropy normalization as a reward function, thus ensuring fairness in weighted energy consumption and delay. What’s more, ENSAC algorithm also considers UAV trajectory planning, task offloading ratio, and power allocation in the UAV-assisted MEC system. Therefore, compared with previous methods, ENSAC algorithm has stronger stability, better exploration performance, and can handle more complex environments and larger action space. Finally, extensive experiments demonstrate that, in both energy-saving and delay-sensitive scenarios, the ENSAC algorithm can quickly converge to the optimal solution while maintaining stability. Compared with four benchmark algorithms, it reduces the total system cost by 52.73%.

The deep clustering algorithm can learn the latent embedded features of the data through the autoencoder, and cluster the data according to the similarity of the latent features. However, the feature information obtained by the autoencoder may not have a better value for the clustering algorithm and is not suitable for clustering, which greatly reduces the clustering effect. This paper proposes a deep K-means clustering algorithm with implicitly embedded space transformation to answer this question. We implicitly transform the latent feature space into a new type of space that is more friendly to the clustering task, which preserves space invariance. This implicit transformation is done through an orthogonal transformation matrix. The orthogonal transformation matrix is composed of the eigenvectors of the intra-class scattering matrix and the inter-class scattering matrix. In the new space, clusters can be better separated by cluster cohesion and inter-cluster difference. We alternately optimize feature acquisition and clustering to adjust the embedding space and disperse the embedding points, to enrich the clustering information in the latent feature space. Experimental results show that our proposed algorithm can produce better high-quality clusters than many current correlation clustering algorithms on the same experimental dataset.

Summary With the rise of industrialization, the importance of the industrial Internet of Things (IIoT) has increased significantly, and with it comes a variety of security threats. Therefore, the security of these networks is critical. Industrial Response Systems (IRSs), as the last line of security, plays an important role in the security system of the Industrial Internet of Things. In this paper, a new IRS model based on the non‐cooperative game is proposed. First, by combining the Partially Observable Markov Decision Process (POMDP) model with the stochastic game model based on the expanded attack tree, our model could effectively perceive the changes at each node. Second, our model incorporates the alarms of intrusion detection system (IDS) and the physical quantities of sensors in Industrial Cyber‐Physical System (ICPS) into the quantization system so that the model can respond to intruders more accurately and comprehensively. Finally, we develop this model based on multiprocessors to speed up the solution process, and adopt an approximation algorithm to reduce the number of iterations of the POMDP

The video moment retrieval task aims to fetch a target moment in an untrimmed video, which best matches the semantics of a sentence query. Existing methods mainly focus on utilizing two separate modules: one learns intra-modal relations to understand video and query contents, and the other explores inter-modal interactions to build a semantic bridge between video and language. However, intra-modal relations information can be easily overlooked when capturing inter-modal interactions. In fact, intra-modal relations and inter-modal interactions can be learned simultaneously within a unified module to make video and sentence guide each other. Towards this end, we propose a Cross-Modal Interaction Network (CMIN) for video moment retrieval by jointly exploring the intra-modal relations and inter-modal interactions between video frames and query words. In CMIN, a query-guided channel attention module is designed to suppress query-irrelevant visual features and enhance crucial contents; then a cross-attention module simultaneously considers intra-modal relations within each modality and fine-grained inter-modal interactions between frames and words, to enhance the semantic relevance between video and sentence query. Compared to the state-of-the-art methods, the experiments on two public datasets (Charades-STA and TACoS) demonstrate the superiority of our method.

Recently, with the large-scale outbreak of the global financial crisis and public safety incidents (such as COVID-19), high-performance computing has been widely applied to risk prediction, vaccine development, and other fields. In scenarios where high-performance computing infrastructure responds to the instantaneous explosion of computing demands, a crucial issue is to provide large-scale flexible allocation and adjustment of computing capability by rapidly constructing computing clusters. Existing large-scale computing cluster deployment solutions usually utilize source code deployment or other deployment tools. The great challenge of existing deployment methods is to reduce excessive image distribution time and refrain from configuration defects. In this article, we design an intelligent distributed registry deployment (IDRD) architecture based on the OpenStack cloud platform, which adaptively places distributed image repositories using the containerized deployment of multiple registries. We propose a server load priority algorithm to solve multiple registries placement problems in IDRD. Furthermore, we devise a clustering algorithm based on demand density that can optimize the global performance of IDRD and improve large-scale cluster load balancing capabilities, which has been implemented in the TianHe Supercomputing environment. Extensive experimental results demonstrate that IDRD can effectively reduce$30\%$-$50\%$of the distribution time of component images and significantly improve the efficiency of large-scale cluster deployment.

Large-scale data centers have been widely used for cloud services, and the stability of various cloud services has received additional attention from users. Although service disruptions are not as catastrophic as they once were, their impact might be more extensive than before. These outages may trigger the migration of virtual machines (VMs) located in the failure node. However, the access time of each VM is random, unlike the accident time, which can be predicted. This means that traditional migration caused by service interruptions may result in a large number of unwanted migrations, regardless of the user’s downtime experience. Migration is an expensive process in terms of the resources needed as well as the degradation of application performance during migration. A balance between the recovery time of the service (to minimize the migration resulting from a given placement) and the downtime experience of the users (to minimize the impact of access interruptions) is needed. In this paper, we propose HMGOWM, a hybrid decision-making mechanism for automating the migration of VMs. Our proposed mechanism extends the original VM migration performance cost model, greatly reducing the downtime experience of the users. To achieve high performance and a good load balance, a multi-objective monitoring system for both VMs and physical machine nodes and an adaptive VM migration-scheduling scheme for the OpenStack cloud platform are proposed. Extensive experiment results indicate that the downtime experienced by users can be efficiently reduced and that the implementation of HMGOWM outperforms the original scheduling of the OpenStack cloud platform.

Nowadays, cloud-supported Internet of Things (Cloud-IoT) has been broadly deployed in smart medical systems, where the limitations of Internet of Things (IoT)-associated medical devices in terms of data access, storage, scalability, and computing are solved through the use of cloud computing architectures. However, with the rapid development of medical equipment and the increasing number of medical devices, it will be extremely difficult to program or manage such an expanding and massive medical IoT system in traditional single-cloud platforms. In this article, we design and implement a multicloud framework for building OpenStack-based platform for medical IoT, referred to as the tri-storage failure recovery system (Tri-SFRS). To implement Tri-SFRS, we combine several techniques to achieve this reduction in effort, including a multicloud cascading architecture, a low-overhead native testing framework, a medical data storage-backup mechanism, and snapshot-volume cascaded operations for b-ultrasonic data. Tri-SFRS is also able to simultaneously enable resource management specialization. Tri-SFRS has been designed as a native component in the OpenStack platform, and it demonstrates the degree of native OpenStack multicloud platform management by our proposed cascading framework. Comparing with the traditional single-cloud OpenStack platform, Tri-SFRS can reduce the resource-request processing latency from B ultrasonic machines by up to 20%. Our experiments also demonstrate the broad applicability of Tri-SFRS.