Real-time data sharing at the IoV edge faces high mobility, bursty delays, and constrained RSUs. We present DBG–LB, an edge-deployable loop that couples Dynamic Behavior Graphs (DBG), a time-weighted trust model (TWDT-Trust), a trust-weighted hybrid BFT (TWH-BFT), and budget-bounded incentives. DBG encodes spatio-temporal interaction quality and suppresses correlated evidence via discounting and adaptive decay. TWDT-Trust then fuses direct observations with credibility-controlled recommendations and penalties. TWH-BFT uses a per-round trust snapshot to drive committee sampling and weighted voting, and it supports asynchronous finalize/rollback under heavy-tailed delays. Finally, budget-bounded incentives settle rewards and penalties on chain and feed back to trust and eligibility. In trace-driven emulations with urban/highway mobility, heavy-tailed delays, and collusion/Sybil/replay attacks, DBG–LB sustains sub-second end-to-end latency (median about 0.6 s) and about 200 transactions/s with low malicious inclusion. Compared with PBFT and a PoW-like proxy, it reduces both latency and energy per transaction while maintaining similar or higher finalization. Ablations confirm the necessity of correlation discounts and adaptive decay. Practically, these properties support city-scale V2X deployment by improving road safety and operational efficiency within realistic edge budgets and enabling auditable, policy-compliant collaboration between fleets and authorities.

The computing power network (CPN) interconnects dispersed heterogeneous computing nodes through a network and achieve rational allocation and efficient utilization of computing power through unified algorithms. Meanwhile, with its decentralized nature, it serves as an ideal platform for decentralized federated learning. However, the mutual distrust among computing providers, computing users, and between computing providers and computing users leads to untrustworthy models in federated learning, and decentralized federated learning also suffers from privacy issues such as data leakage, hindering its application in CPN. Therefore, to address the aforementioned problems, we propose a decentralized and verifiable federated learning approach for computing power network, termed DVFL-CPN. Specifically, we first propose a single-masking encryption algorithm based on pseudorandom numbers, primarily to protect the local models trained by each computing node. Then, we propose a verification algorithm based on homomorphic hashing to verify the aggregated results of computing nodes. Subsequently, we propose a robust aggregation algorithm, enabling online nodes to complete model aggregation and decryption even when some computing nodes go offline. We also deploy a blockchain to record the operation results of each node. Finally, we conduct experiments on real-world datasets. Compared to state-of-the-art decentralized and verifiable federated learning methods, our method achieves a 13.5% improvement in accuracy and demonstrates robustness against offline nodes. Additionally, our method outperforms existing methods in terms of computational efficiency and communication efficiency by approximately 3×104and 1.6×105times, respectively.

The increasing use of wireless sensor networks (WSNs) necessitates rapid detection and mitigation of system anomalies and state changes, which can be achieved through change point detection (CPD) methods. This article introduces a novel approach to detect change points in WSNs, employing an integrated multimodal method that combines three innovative feature extraction models and a learnable weighted fusion layer for evidence synthesis. Leveraging subjective logic and Dempster-Shafer theory with multivariate time series deep learning, the proposed approach establishes a prior probability distribution informed by subjective logistic loss, thus enhancing decision-making reliability by quantifying uncertainty. To address sample imbalance, this study integrates subjective logistic loss with a sample size parameter and introduces a Kullback-Leibler (KL) loss to prevent overconfident errors. A novel semi-supervised training model employing self-distillation and a multimodel KL loss is also proposed, which significantly improves accuracy and robustness. Comprehensive experiments validate the method, with accuracy improvements of up to 97.79% and 99.69% on various datasets, setting a new benchmark for CPD performance and demonstrating the method’s potential for real-world applications.

In-band network telemetry (INT) is a new network measurement technique that provides real-time, fine-grained packet-level network measurements. However, standard INT lacks the flexibility to perform configurable on-demand network measurements. In this work, we propose a flexible on-demand network measurement mechanism (FDSR-INT) based on a dual bitmap by combining the programmability of segment routing based on IPv6 (SRv6) and the telemetry efficiency of the INT to achieve customizable network measurements. By designing the dual bitmap, the telemetry information of personalized probe nodes is supported, and the telemetry efficiency is improved. SRv6 is employed to direct measurement probe packets, ensuring coverage of a specified set of network targets. Its inherent programmability enables INT to conduct customized per-hop network measurements. The designed flexible INT field structure, which appends traceability information along with telemetry information, reduces the information traceability overhead in the control plane and improves the network compatibility of the mechanism. We solve the optimal probing path by solving a traveler problem in an auxiliary graph and design a greedy path-cutting algorithm to maximize the number of nodes for single packet probing while satisfying the packet length constraints to improve the success rate of the probing task. Finally, we implemented FDSR-INT using P4 and verified its performance experimentally in the constructed space-air–ground integrated simulation dynamic environment. FDSR-INT saves 30% of the data plane bandwidth and 54% of the data plane bandwidth on north-south interfaces compared with SONM-SR-INT, etc. Furthermore, it has low control plane processing overhead and probe path transmission overhead.

The purpose of malicious traffic detection and identification in the Internet of Things (IoT) is to detect the intrusion of malicious traffic within the IoT network into IoT devices. Detection and identification play a key role in ensuring the security of the IoT. At this time, great success has been achieved with deep learning in the field of malicious traffic detection and identification. However, due to resource limitations, such as computation weaknesses and low-edge network node storage capacity in the IoT, a high-complexity model based on deep learning cannot be deployed and applied. In this article, we propose a lightweight malicious traffic detection and recognition model named lightweight knowledge distillation space time neural network (LKD-STNN) based on knowledge distillation (KD) deep learning for the IoT. We use KD to build a lightweight student model by depthwise separable convolution and bidirectional long short-term memory (BiLSTM) to realize a lightweight student model and obtain multidimensional characteristic information. According to the characteristics of KD, we propose an adaptive temperature function that can adaptively and dynamically change the temperature during the process of knowledge transfer so that different softening characteristics can be obtained during the training process. Then, the weight is updated by combining loss functions to improve the performance of the student model. The experimental results show that with the publicly available malicious traffic data sets for the IoT, the ToN- IoT and IoT-23, our model not only reduces the complexity of the model and the number of model parameters to less than 1% of the teacher model but also reaches an accuracy of more than 98%, indicating that our model can be applied to the multiclassification identification of malicious traffic in the IoT.

Abstract Table-to-text generation aims to generate descriptions for structured data (i.e., tables) and has been applied in many fields like question-answering systems and search engines. Current approaches mostly use neural language models to learn alignment between output and input based on the attention mechanisms, which are still flawed by the gradual weakening of attention when processing long texts and the inability to utilize the records’ structural information. To solve these problems, we propose a novel generative model SAN-T2T, which consists of a field-content selective encoder and a descriptive decoder, connected with a selective attention network. In the encoding phase, the table’s structure is integrated into its field representation, and a content selector with self-aligned gates is applied to take advantage of the fact that different records can determine each other’s importance. In the decoding phase, the content selector’s semantic information enhances the alignment between description and records, and a featured copy mechanism is applied to solve the rare word problem. Experiments on WikiBio and WeatherGov datasets show that SAN-T2T outperforms the baselines by a large margin, and the content selector indeed improves the model’s performance.

The existing personalized text-to-image generation models face issues such as repeated training and insufficient generalization capabilities. We present an adaptive Style-Guided Diffusion Model (SGDM). When provided with a set of stylistically consistent images and prompts as inputs, SGDM can generate images that align with the prompts while maintaining style consistency with the input images. SGDM first extracts features from the input style image and then combines style features from different depths. Last, style features are injected into the noise generation process of the original Stable Diffusion (SD) model by the style-guided module we propose. This strategy fully leverages the generative and generalization capabilities of the pre-trained text-to-image model to ensure the accuracy of the generated image's content. We present a dataset construction method suitable for style personalized generation tasks of this kind, enabling the trained model to generate stylized images adaptively instead of re-training for each style. We also present an evaluation metric, StySim, to measure the style similarity between two images, and this metric shows that the style personalization capability of SGDM is the best. And metrics such as FID, KID, and CLIPSIM indicate that SGDM maintains good performance in text-to-image generation.

At present, deep learning technologies have been widely used in the field of natural language process, such as text summarization. In CQA, the answer summary could help users get a complete answer quickly. There are still some problems with the current answer summary scheme, such as semantic inconsistency, repetition of words, etc. In order to solve this, we propose a novel scheme Answer Summarization based on Multi-layer Attention Scheme (ASMAM). Based on the traditional Seq2Seq, we introduce self-attention and multi-head attention scheme respectively during sentence and text encoding, which could improve text representation ability of the model. In order to solve "long distance dependence" of RNN and too many parameters of LSTM, we all use GRU as the neuron at the encoder and decoder sides. Experiments over the Yahoo! Answers dataset demonstrate that the coherence and fluency of the generated summary are all superior to the benchmark model in ROUGE evaluation system.

As the emerging technology of Unmanned Aerial Vehicles (UAVs) becomes mature, UAVs are widely used in environmental monitoring, communication and other fields. In view of this, this paper analyzed the task of synergistic multi-UAVs exploration of unknown environments, and proposed a path planning method for them based on reinforcement learning. Firstly, the path planning task of the UAVs was divided into two parts: the path travel strategy module and the information exploration strategy module. Models of the two modules were based on the Deep Deterministic Policy Gradient algorithm (DDPG), and an improved Artificial Potential Field (APF) force traction mechanism was introduced in the path travel strategy module. Its aim was to assist in guiding the generation of UAV flight path trajectories. Also it could enhance the learning capability of the model. The path travel strategy module would generate the complete flight path of the whole cluster in a distributed manner. A series of temporary target points provided by the information exploration strategy module helped. In maps with 21.5%, 25.3% and 29.6% of obstacles, multi-UAVs could achieve 84.2%, 76.7% and 69.9% of environmental exploration by the designed method. Compared with the APF method, the A star method and the Breath First Search (BFS) method, the proposed method is not only able to plan feasible paths in a more complex map model, but also the curvature of the planned paths is smoother, thus achieving the goal of reducing the energy cost of UAVs.

With the continuous development of artificial intelligence technology, machine learning in distributed network systems, such as IoVflntemet of Vehicles), will inevitably lead to privacy leakage. At present, there are lots of problems in federated learning, differential privacy and other machine learning privacy protection schemes, such as high loss of availability and the need of a fully trusted third-party server. In order to solve these problems, we propose MDPFL(Multiple Differential Privacy based on Federated Learning) algorithm. The algorithm combines with the differential privacy model in each stage of federated learning to solve the problem of curious third-party data collectors obtaining users’ original data in the process of machine learning. Meanwhile, the algorithm does not adopt a decentralized machine learning scheme directly, but uses a double noise adding mode with the existence of the third-party data collectors. We use Laplacian mechanism in the central server and GRR mechanism in local clients to ensure the goal of stability of the effect of the machine ¡earning model. The algorithm is compared with FedAvg, CDP, LDP algorithm in accuracy and loss rate which based on EMNIST data sets. In different global sensitivities, the model training effect is consistent with FedAvg algorithm while comparing with LDP algorithm, the speed of model convergence is improved.

Syslog records critical information of network when the system is running, and has been used to help practitioners carry out various network maintenance and operation activities. Because of abundance of syslog, automated log analysis technology is needed to complete the above activities. Usually, log parsing is an important part of log analysis. Previous log parsing methods have good achievements, but they heavily rely on well-designed regular expressions and ignore semantic information of log. To solve these problems, we propose a novel log parsing method SNNLog, which uses Siamese Network to assess the similarity between log messages, and the similarity is used for the subsequent parsing. In addition, after parsing, SNNLog merges log similar events, reducing the misjudgment rate caused by non-numeric token variables at the beginning of messages. We evaluated on five publicly accessible log datasets and the results show that compared with SOTA algorithms such as LogMine, Spell, Drain and MoLFI, SNNLog has an F1-score of 0.999 on five datasets, and parsing accuracy of four datasets is the highest.

To cope with the task of malicious traffic detection in high-complexity and high-security-demands industrial internet of things scenarios, this paper presents a new malicious traffic detection model deployable at edge computing nodes in the industrial internet of things scenarios along with the training method by combining the multi-head self-attention mechanism with neural network architecture partial search for the first time in this field along with a new, generalized linear multidimensional projection method. In addition, this paper presents the partially learnable embedding based on two-dimensional Gaussian distribution for the first time to capture the absolute position information. The experimental results show that this model is both lightweight and efficient, and the accuracy is much higher than previous studies with the same dataset.

With the increase of car ownership in China, the problem of difficult parking in cities has become more and more serious. In large parking garage, finding an ideal parking space has become a daily problem for people; especially during the peak usage period of parking garage, a large number of vehicles will be driven into the parking garage, which will bring congestion, and at the same time, the demand of a large number of users in the area will lead to network congestion under 4G network, and users cannot access the network to check and find parking spaces. Therefore, this paper proposes a global optimal navigation mechanism with 5G network based on the above mentioned problems by analyzing the congestion situation during the peak usage of the parking garage. This mechanism uses an optimal parking space selection model and a global optimal scheduling model that can avoid congestion. Based on the model proposed in this paper, we construct a navigation planning system based on mobile, front-end and back-end. With the ultra-low latency, ultrahigh transmission efficiency and reliability of 5G network, the mobile port can select the optimal parking space according to the user's preference, and the back-end can show the parking space and congestion in the parking garage in real time according to the actual situation of the parking garage. Finally, the performance of the model and system proposed in this paper is verified through experiments.

Disk failure is one of the most important reliability problems in large-scale network storage systems. Disk failure may lead to serious data loss and even disastrous consequences if the missing data cannot be recovered. Therefore, predicting disk failures is an important means of ensuring storage security in network storage systems. However, since the fault data in the fast degradation stage is smaller than the healthy data in the normal state, the mixture of healthy data and faulty data leads to extremely unbalanced data, which brings great challenges to finding hidden fault information, thus making fault prediction more accurate. Aiming at the above problems, a convolutional transformer model ConvTrans-TPS model for disk failure prediction in large-scale network storage systems is proposed. The ConvTrans-TPS model acquires dependencies between long-term sequence data through transformers and uses convolutional projections for attention computation to enhance attention to local contextual information. Data augmentation to predict failures in the next 7 days. Validated by the analysis on the Backblaze dataset, the F1 is 0.96 and the Matthews correlation coefficient (MCC) is 0.92. Compared with the popular CNN-LSTM model in recent years, our proposed method improves F1 and MCC by 4% and 5%, respectively, improving the prediction accuracy.

With the increasing types and number of Internet of Things (IoT) devices and malicious programs and the popularization of encryption technology in the communication process between the Internet and the IoT, a large amount of encrypted abnormal traffic among devices endangers IoT cybersecurity. How to identify abnormal encrypted traffic of the IoT has become the premise of cybersecurity. Presently, most of the detection methods for traffic in the IoT have problems, such as simple data set processing, imperfect feature extraction, data imbalance, and low multiclassification accuracy. In this article, we propose a multiclassification deep learning model named the cost matrix time–space neural network (CMTSNN) for abnormal and encrypted IoT traffic. The CMTSNN is divided into three parts. The first part is the preprocessing stage of the data set, which needs to retain the timing relation between two data packets in the stream and create a cost penalty matrix according to the sample distribution. Aimed at the robustness of feature extraction in network flow, the second part extracts time series features and then space features to ensure the robustness of feature extraction. The third part is aimed at the problem of data imbalance. The cost penalty matrix is applied to the cost penalty layer in the training process, and then the improved cross-entropy loss function is used to calculate the loss to improve the classification accuracy of minority categories and increase the overall multiclassification performance of the model. Experiments were carried out with the ToN-IoT, BoT-IoT, and ISCX VPN-NonVPN data sets. Compared with current methods, the proposed method shows better performances, including accuracy, precision, recall, F1 Score, and false alarm rate.

Electrocardiogram (ECG) is a common diagnostic indicator of heart disease in hospitals. Because of the low price and noninvasiveness of ECG diagnosis, it is widely used for prescreening and physical examination of heart diseases. In several studies on ECG analysis, only rough diagnoses are made to determine whether ECGs are abnormal or on a few kinds of ECG. In actual scenarios, doctors must analyze ECG samples in detail, which is a multilabel classification problem. Herein, we propose Hygeia, a multilabel deep learning-based ECG classification method that can analyze and classify 55 types of ECG. First, a guidance model is constructed to transform the multilabel classification problem into multiple interrelated two-classification models. This method ensures the good performance of each ECG analysis model, and the relationship between various types of ECG can be used in the analysis. The imbalance of samples in ECG datasets makes it difficult to analyze abnormal ECGs with high sensitivity and accuracy. We used data generation and mixed sampling methods for 11 ECG types with imbalanced problems to improve the average accuracy, sensitivity, F1 value, and accuracy from 87.74%, 43.11%, 0.3929, and 0.3929, to 92.68%, 96.92, 0.9287, and 99.47%, respectively. The average accuracy, sensitivity, F1 value, and accuracy of 44 of the 55 tags of the abnormal ECG analysis model are 99.69%, 95.81%, 0.9758, and 99.72%, respectively.

The existing solutions related to local differential privacy (LDP) in multi-layer networks for edge computing scenarios present several limitations in both key-value data heavy hitter identification and related frequency and mean estimation tasks. First, existing LDP approaches cannot effectively use edge nodes to improve their utility/performance. Secondly, there are many network transmission tasks in edge computing, which have relatively high requirements for communication and storage costs. Furthermore, the traditional privacy budget allocation cannot attain the best utilization. To solve the above problems, we propose MLPKV, a local differential multi-layer private key-value data collection scheme for edge computing, structured into three phases: dimensional reduction, padding-length estimation, and estimation. An improved EC-OLH algorithm is used to offload the computing efforts related to aggregation and estimation to edge nodes for achieving greater efficiency. In the dimensional reduction phase, a candidate set is generated to prune the domain of original data, which improves the estimation. In addition, our method groups users for completing the tasks in each phase to avoid additional errors caused by dividing the privacy budget, and proposes a new user division with an optimal grouping ratio. Finally, the proposed method was implemented in a proof-of-concept prototype system. We compare MLPKV with baseline methods such as PrivKV and PCKV. Experimental results on both synthetic and real-world datasets show that our method achieves better utility for heavy hitter identification, frequency, and mean estimations than other state-of-the-art mechanisms. For small data sets, our approach also provides high-accuracy estimation with a low privacy budget.

Blockchain can help edge IoT-enabled Maritime Transportation Systems (MTS) in solving its privacy and security problems. In this paper, a lightweight blockchain called LBlockchainE is designed for edge IoT-enabled MTS to guarantee the security of sensor data stored in an edge computing environment. To save the resources of edge servers on ship, a data placement strategy is proposed. To encourage edge servers to positively contribute to storing data generated by sensor devices, storage resource consumption is employed as an influencing parameter, and servers with abundant resources are selected for priority storage. The data placement strategy also takes care of the access delay between servers and selects the nodes with the least access and storage costs as the priority storage choice. LBlockchainE applies the low-energy-consumption characteristics of Proof of Stake to determine the ownership of bookkeeping rights through a small number of competitive calculations and the resources of the node. Experimental results indicate that compared with Ethereum, the consensus mechanism of LBlockchainE consumes less energy and occupies less storage space. On average, the new system uses 1.6% less time and consumes 78% less battery power compared with traditional blockchain systems. In comparison to the random storage, the best storage, and the optimal data storage strategies, the proposed strategy maintains the same message costs.

Many end-to-end deep learning algorithms seeking to classify malicious traffic and encrypted traffic have been proposed in recent years. End-to-end deep learning algorithms require a large number of samples to train a model. However, it is hard for existing methods fully utilizing the heterogeneous multimodal input. To this end, we propose the SHAPE model (simultaneous header and payload encoding), which mainly consists of two autoencoders and a transformer layer, to improve model performance. The two auto encoders extract features from heterogeneous inputs—the statistical information of each packet and byte-form payloads—and convert them into a unified format; then, a lightweight Transformers layer further extracts the relationship hidden in simultaneous input. In particular, the autoencoder for payload feature extraction contains several depthwise separable residual convolution layers for efficient feature extraction and a token squeeze layer to reduce the computing overhead of the Transformers layer. Moreover, we train the SHAPE model using deep metric learning, which pulls samples with the same class label together and separates samples from different classes in the low-dimensional embedding space. Thus, the SHAPE model can naturally handle multitask classification, and its performance is approximately 5.43% better than the current SOTA on the traffic type classification of the ISCX-VPN2016 dataset, at the cost of 9.31 times the training time, and 1.45 times the inference time.

The development of the mobile Internet has led to a shape increase in video traffic data. New Mobile Edge Computing (MEC) technology can reduce network operation and service delivery delays, so as to improving the QoE (Quality of Experience) of user. In MEC environment, this article builds a simulation platform for multi-bitrate video caching and scheduling scenarios VECSim (Video Edge CloudSim). VECSim provides a simulation environment specific to verify multi-bitrate video caching collaboratively and scheduling strategy in the Mobile Edge Computing scenarios. Unlike simple numerical simulations and complex real-world deployments, VECSim provides a simple interface that can be called by the user, besides, it can be used to provide larger-scale, more flexible request scenarios which are closer to the actual environment. Based on the edge computing simulation platform EdgeCloudSim, this platform adds and modifies quite a few functions so that it can be effectively used in multi-bitrate video caching and scheduling scenarios, which now includes seven main modules: Server Management Modules, Network Module, User Mobility Module, Task Loader Generator Module, Video Storage Module, Edge Client Management Module, and Edge Orchestrator Module. With this platform, researchers can easily and quickly verify their video caching and scheduling strategies.

Relation classification aims to classify the entity pairs into a certain relation, which is an important task of natural language processing. The latest end-to-end models based on attention mechanism still have shortcomings, i.e., the attention will be gradually weakened when processing long sequences, and they cannot make use of the hidden type information of the entities. To solve these problems, we propose a relation classification model based on the selective entity-aware attention mechanism, which consists of context encoder and entity-aware attention network. In the context encoder, contextual word semantics are learned through self-attention. Entity selection is applied to adapt the fact that different words can determine each other’s importance. Latent types of entities are taken as auxiliary information to make full use of the entities’ hidden features. Experiments on the SemEval-2010 Task 8 dataset and TACRED show that our model outperforms the baselines without implementing any external resources or NLP tools, and the entity-aware attention indeed improve the model’s performance.

Multimodal sentiment classification is an important research attracting many scientists’ attention in natural language processing. In most multimodal sentiment research, each modal of the dataset is labeled with a unified label. However, this unified label of multimodal data may limit the model to obtain the different information between multimodal in the training process. To address the above issues, this paper proposes AtM-DNN, a model based on multimodal attention fusion network in collaboration with auxiliary function for multimodal sentiment classification. Different single-modal encoders are used to map the original data of different modal into vectors of different information dimensions. Attention mechanism is introduced to control the information weight of each modal in the interaction, so as to remove the redundancy of information. Besides, each modal is introduced into the loss function as an auxiliary function to assist the calculation of the final loss function. Experiments on CH-SIMS dataset show that the attention mechanism of each modal can exactly improve the accuracy and F1 of sentiment classification of each single-modal, and also can improve the accuracy and F1 of multimodal sentiment classification. Although the collaboration of auxiliary function can improve the effect of multimodal sentiment classification, it is likely to reduce the effect of each single-modal. And these models are better than the single-modal sentiment classification model.

Encrypted network traffic classification is an essential task in modern communications, which is used in a wide range of applications, such as network resource allocation, QoS (Quality of Service), malicious detection, etc. With the continuous evolution of network technology, approaches used to classify network encrypted traffic have become increasingly complex, and model training relies heavily on a large amount of labeled data. However, the acquisition of correct and massive labeled data of network traffic in the real environment remains a major challenge in this field. On the opposite, unlabeled traffic is extremely easy to obtain in a network. Therefore, the effective use of unlabeled data is of great significance to the development of modern communications. In this paper, we propose the Sauce model, which can effectively use unlabeled information to obtain high-quality clustering space. It is composed of Aux (auxiliary network) and AE (auto encoder), where Aux is designed for collaborative training with AE, affecting the distribution of samples in the latent space generated by AE. Sauce uses t-SNE (t- distributed stochastic neighbor embedding) to perform secondary dimensionality reduction on the latent code, and then uses a clustering algorithm for cluster analysis. Besides, sauce applies the self-training technique, replacing the real labeled data with pseudo-labeled data to reduce the reliance on labeled data and improve the utilization of unannotated data. We conduct experiments on two real network datasets. The experiments show that Sauce can achieve clustering accuracies of 98.4% (VPN dataset) and 98.0% (TOR dataset), outperforming other unsupervised or semi-supervised learning methods.

Effective workload forecasting can provide a reference for resource scheduling in cloud data centers. Compared with the normal single data center, the multi-data center has a more complicated architecture design and provides more diverse computing services. The traditional load forecasting models need too much manual intervention to set up the parameters, meanwhile, the emerging neural network methods are not sensitive to the prediction scale and cannot capture the long-term associations between input features. To address these challenges, we propose a hybrid model named Attentions-based LSTM Encoder-Decoder network and Autoregressive model (ALEDAR), which combines neural network and statistical learning methods, to analyze the linear and nonlinear characteristics of the load sequence over time in a multi-cloud data center environment. ALEDAR uses a dual attention-based Encoder-Decoder architecture to extract the relationships between historical workload data, the design can also avoid the deterioration of prediction effect caused by long-range data scale, the output layer is composed of a three-layer perceptron. Moreover, ALEDAR employs an autoregressive module to capture the linear trend of the load sequence and eliminate scale insensitivity of input and output data. The experimental results show that our approach is adaptive and can improve the performance of host workload prediction in both single cloud data center and multi-cloud data center environments. We evaluate the performance of the proposed algorithm on real-world data sets and observe consistent improvement of 9.7% - 34.2% over the state-of-the-art baselines.

In order to verify the effectiveness of task offloading algorithms in large-scale cloud and edge computing systems, scholars usually use simulation platforms to conduct extensive experiments. However, existing simulation platforms have unclear application scenarios and relatively homogeneous functions, which cannot fully exploit the respective advantages of cloud and edge computing. To address the above issues, this paper designs a new simulation platform, CETO-Sim (Cloud-Edge Task Offloading Simulation), for the cloud-edge aggregation computing scenario, which is convenient for users to verify the effectiveness of the task offloading algorithm. The platform adopts a modular design strategy, allowing users to invoke the various interfaces provided by the platform to simulate and build different physical entities, and to freely combine them to simulate different computing environments. The platform contains six components: task management component, network topology component, end-user component, scheduling policy component, server component and operation component. The platform simulates the task offloading process through the interaction between these components, while recording the changes of various parameters of tasks and simulated entities in real time, and finally printing out the experimental results in the form of logs. This paper presents an example of task offloading using the particle swarm algorithm in a cloud-side aggregated computing environment, demonstrating that the CETO-Sim platform has the advantages of clear architecture, simple operation, easy deployment, accurate experimental results and high simulation efficiency, and also verifying the feasibility of deploying the platform in other scenarios.

Existing navigation services provide route options based on a single metric without considering user’s preference. This results in the planned route not meeting the actual needs of users. In this paper, a personalized route planning algorithm is proposed, which can provide users with a route that meets their requirements. Based on the multiple properties of the road, the Polychromatic Sets (PS) theory is introduced into route planning. Firstly, a road properties description scheme based on the PS theory was proposed. By using this scheme, users’ travel preferences can be quantified, and then personalized property combination schemes can be constructed according to these properties. Secondly, the idea of setting priority for road segments was utilized. Based on a user’s travel preference, all the property combination schemes can be prioritized at relevant levels. Finally, based on the priority level, an efficient path planning scheme was proposed, in which priority is given to the highest road segments in the target direction. In addition, the system can constantly obtain real-time road information through mobile terminals, update road properties, and provide other users with more accurate road information and navigation services, so as to avoid crowded road segments without excessively increasing time consumption. Experiment results show that our algorithm can realize personalized route planning services without significantly increasing the travel time and distance. In addition, source code of the algorithm has been uploaded on GitHub for this algorithm to be used by other researchers.

In medical scenarios supported by edge clouds, it is difficult for patients to truly gain ownership of their electronic health records (EHRs). However, it is easy for doctors to modify hospital data to deny incorrect treatment records, which makes it difficult to protect the rights of patients. To improve patient control over EHRs, an attribute-based encryption protection scheme named CEC-ABE for EHRs combined with a blockchain is proposed to protect EHRs in edge cloud environments. In this scheme, the agreement process between the patient and the hospital is completed before the ABE stage, and the treatment information, including the treatment time, treatment doctor and additional information, is confidentially transmitted through an encryption algorithm. By storing the uploaded encrypted data in the blockchain in the form of transaction records, the integrity of the data is guaranteed, which facilitates the traceability of EHR generation. Access to EHRs is controlled by the ABE scheme of the outsourced ciphertext policy, and fine-grained attribute revocation can be employed to ensure the security of the ciphertext. The CEC-ABE algorithm, CP-ABE algorithm and other algorithms are experimentally tested, and the computational cost of each stage of the algorithms and the computational delay of each role are compared. CEC-ABE can significantly improve performance in key generation, outsourced decryption and other stages. Compared with the algorithm whose performance is second only to CEC-ABE, it reduces the computational overhead by 1.73% and 5.2%. The results show that the overall comprehensive performance of the CEC-ABE algorithm is better than that of the other algorithms.

Intrusion detection system (IDS), as a network security device, monitors network data in real time and responds actively when it detects suspicious transmissions. However, suffered from the large amount of redundancy and high correlation of network data, the traditional IDS have defects in low detection rate and high computational overhead. In this paper, we propose a network data classification mechanism (CPEL). Data preprocessing of network traffic data is first performed using correlation-based feature selection (CFS) and principal component analysis (PCA). CFS selects the best features in the data, PCA to dimensionality reduction and denoising. Then, we use multiple classifiers to perform anomaly detection, and select the best three classifiers according to the classification effect. On this basis, we use the majority vote as the ensemble learning (EL) method of combination rules to further improve the performance of the classifier. Experimental results on the benchmark NSL-KDD and WSN-DS datasets indicate that this mechanism outperforms a single algorithm in accuracy and detection rate. In the meantime, computational overhead is remarkably reduced.

Community Question Answering (CQA) provides a platform to share knowledge for users. With the increasing number of users and questions, askers have to wait a long time for an answer with high quality while responders may not be interested in assigned questions. Current methods usually try to address this issue based on text or link analysis. However, most of them suffer from delayed answers or low coverage of best answer. In this paper, we design a novel expert recommendation mechanism by incorporating the deep structured semantic model (DSSM) [20] with our proposed graph-based algorithm, a topic sensitive answerer rank algorithm (TSAR). In the process of constructing transition probability matrix, we not only take into account both the number of questions answered by the user and question difficulty, but also consider the user's average response time for providing the answer. The experiments carried out on Yahoo! Answers and Stack Overflow datasets demonstrate that the proposed mechanism outperforms the current typical algorithms [9] on multiple metrics and achieves the best answer coverages, which are 61.5% and 53.8%, respectively.

In order to select a composition scheme that meets user's needs and high performance from large-scale web services in the edge cloud, this paper proposes a trusted service composition optimization scheme called TSC-ECFA for edge cloud, which applies the predation strategy to the firefly algorithm (FA) and divides the services with the same input and output into one category, reducing the number of combinations to search for feasible solutions. The cotangent chaos theory is used to generate the initial firefly population and disturb the location of the individual to improve the overall search efficiency of FA. This paper also improves the step factor and attractiveness formula of FA to give full play to the detection ability of the step factor in the early stage of the algorithm, and balance the local search and the global search. Considering that the QoS attribute values of the service are vulnerable to be tampered with, the blockchain is used to store the QoS attribute values to ensure the tamper-proof and reliability of the QoS attribute value. Finally, simulation experiments compare the number of iterations and optimization of the three algorithms. The experimental results show that the comprehensive performance of TSC-ECFA is better than other algorithms.

Internet of Things (IoT) refers to the complex systems generated by the interconnections among widely available objects. Such interactions generate large networks, whose complexity needs to be addressed to provide suitable computationally efficient approaches. In this article, we propose a distributed local community detection algorithm based on specific properties of community center expansions (DLCD-CCE) for large-scale complex networks. The algorithm is evaluated via a prototype system, based on Spark, to verify its accuracy and scalability. The results demonstrate that compared to the typical local community detection algorithms, DLCD-CCE has better accuracy, stability, and scalability, and effectively overcomes the problem that existing algorithms are sensitive to the location of initial seeds.

Abstract The dynamic travelling salesman problem (DTSP) is a natural extension of the standard travelling salesman problem, and it has attracted significant interest in recent years due to is practical applications. In this article, we propose an efficient solution for DTSP, based on a genetic algorithm (GA), and on the one-by-one revision of two sides (GORTS). More specifically, GORTS combines the global search ability of GA with the fast convergence feature of the method of one-by-one revision of two sides, in order to find the optimal solution in a short time. An experimental platform was designed to evaluate the performance of GORTS with TSPLIB. The experimental results show that the efficiency of GORTS compares favourably against other popular heuristic algorithms for DTSP. In particular, a prototype logistics system based on GORTS for a supermarket with an online map was designed and implemented. It was shown that this can provide optimised goods distribution routes for delivery staff, while considering real-time traffic information.

Industry 4.0 has drawn considerable attention from industry and academic research communities. The recent advances in Internet of Things (IoT), Big Data analytics, sensor technology, and artificial intelligence have led to the design and implementation of novel approaches to take full advantage of data-driven solutions applicable to Industry 4.0. With the availability of large datasets, it has become crucially important to identify the appropriate amount of relevant information, which would optimize the overall analysis of the corresponding systems. In this article, specific properties of dynamically evolving data systems are introduced and investigated, which provide framework to assess the appropriate amount of representative information.

The Massive Open Online Course (MOOC) is based on the Connectionism Theory and the Open Pedagogy of Network Learning, realizing the large-scale open sharing of high-quality education resources via the Internet. MOOC is conducive to the multi-dimension and personalized teaching. However, compared with the mode of traditional classroom teaching, there are also insurmountable drawbacks of MOOC to be addressed. Especially, it is difficult for educators to know the true states of students in real-time when they are learning online, and control the quality of teaching and learning. In this paper, we carried on the thorough research to the arousal degree of learner's brain, and proposed the MOOC-oriented EEG-based arousal of brain detection and adjustment scheme, and constructed the prototype system named Megrez. Megrez can analyze the brain arousal with the EEG data of learners wearing portable EEG detection devices. The energy value of β wave of EEG is selected as the main feature of brain arousal. After preprocessing the EEG data, the SVM classifier is trained, which classification performance is proved satisfactory. When a learner's brain is in a low arousal state, Megrez automatically plays the appropriate music to improve the arousal of brain. The experimental results show that Megrez can effectively detect the arousal of learner's brain, help learners avoid inefficient learning in the negative state, and can also adjust the state of the brain if necessary.

Author name disambiguation is a very important and complex research topic. During the retrieval and research of literature the quality of the investigation results has been reduced because of the high probability of different authors sharing the same name, which lengthens the whole cycle of the scientific research. Therefore, it is necessary to find a reasonable and efficient method to distinguish the different authors who share the same name. In this paper, an author name disambiguation algorithm based on the fusion of multiple features (NDFMF) is proposed. First we proposed a single feature similarity detection algorithm (SFSD). SFSD is used to compute the degree of similarity between two features of a paper and to assess the threshold value. Then, SFSD is used to realize the preliminary SFSD-based disambiguation algorithm (SFSDD). Furthermore, different features are evaluated according to the disambiguation results of author names and the evaluation metrics, including precision, recall and F-measure with SFSDD. The evaluation parameter of weight (W) is introduced to express each feature's influence in disambiguation. NDFMF can disambiguate author names more efficiently based on the fusion of multiple features. Experiments were implemented to test the performance of NDFMF. Experimental results show that NDFMF was effective in the disambiguation precision, recall and F-measure.

This Innovative Practice Work in Progress Paper presents a multi-dimension training scheme to improve the innovation capacity of postgraduate and undergraduate students collaboratively. The current training of postgraduate and undergraduate education schemes cannot form a mutually beneficial relation, not suitable for the training of innovation capacity on postgraduate and undergraduate students collaboratively. There are various problems that need to be addressed. The main strategy of the proposed scheme is utilizing real research programs to establish an integrated team mixed with both postgraduate and undergraduate students, focusing on stimulating their innovation capacity in scientific research. In the scheme, we designed a multi-level joint training mechanism based on the mixed pedagogy, the collaborative resource pool platform, the multi-role and team-based innovation talent incubation mechanism, and the preparatory postgraduate student selection and training goal achievement evaluation mechanism. We applied the scheme to the training process of postgraduate students from the majors of Software Engineering, Computer Software and Theory, Computer Application, Computer Technology and Electronic Information Engineering, and undergraduate students from the majors of Computer Science and Technology, Information Security and Communication Engineering in Nanjing University of Posts and Telecommunications. The training results approve the feasibility and effectiveness of the scheme.

To reduce the operating cost, leasing appropriate amount of public resources becomes a popular practice among small and medium sized enterprises. Many hybrid cloud workflow management systems (HCWMSs) have been developed to provision applications on both local and rented resources. One of the critical issues in the HCWMS is the dynamic resource allocation for stochastically arriving requests. Therefore, we propose a dynamic interval scheduling based heuristic for the resource allocation problem, in which stochastic requests are taken as a set of linearly dependent tasks and distributed to idle and feasible time slots on multiple virtual machines (VMs), either local or rented VMs. The objective is to minimize the idle time slots on the rented VMs, which is relative to the renting cost of VMs, especially for the on-demand pricing structure. Requests arrive at the same time are taken as a batch of tasks to schedule. Tasks are scheduled batch by batch, obeying the precedence constraint and the deadline constraint. We develop a fast heuristic integrated with an interval scheduling to obtain feasible and effective solutions. Three interval scheduling method are proposed and compared: Max Interval Number Scheduling (MINS), Max Working Time Scheduling (MWTS) and Select-the-better Method (STBM). The experimental results show that the interval scheduling based heuristic can reduces the cost of renting VMs.

In recent years, different kinds of natural hazards or man-made disasters happened that were diversified and difficult to control with heavy casualties. In this paper, we focus on the rapid and systematic evacuation of large-scale densities of people after disasters to reduce loss in an effective manner. The optimal evacuation planning is a key challenge and becomes a hotspot of research and development. We design our system based on an Internet of Things (IoT) scenario that utilizes a mobile cloud computing platform in order to develop the crowd lives oriented track and help optimization system (CLOTHO). CLOTHO is an evacuation planning system for large-scale densities of people in disasters. It includes the mobile terminal (IoT side) for data collection and the cloud backend system for storage and analytics. We build our solution upon a typical IoT/fog disaster management scenario and we propose an IoT application based on an evacuation planning algorithm that uses the artificial potential field (APF), which is the core of CLOTHO. APF is conceptualized as an IoT service, and can determine the direction of evacuation automatically according to the gradient direction of the potential field, suitable for rapid evacuation of large population. Based on APF, we propose an evacuation planning algorithm names as APF with relationship attraction (APF-RA). APF-RA guides the evacuees with relationship to move to the same shelter as much as possible, to calm evacuees and realize a more humanitarian evacuation. The experimental results show that CLOTHO (using APF and APF-RA) can effectively improve convergence rate, shorten the evacuation route length and evacuation time, and make the remaining capacity of the surrounding shelters well balanced.

Combinatorial optimization problems are typically NP-hard, due to their intrinsic complexity. In this paper, we propose a novel chaotic particle swarm optimization algorithm (CS-PSO), which combines the chaos search method with the particle swarm optimization algorithm (PSO) for solving combinatorial optimization problems. In particular, in the initialization phase, the priori knowledge of the combination optimization problem is used to optimize the initial particles. According to the properties of the combination optimization problem, suitable classification algorithms are implemented to group similar items into categories, thus reducing the number of combinations. This enables a more efficient enumeration of all combination schemes and optimize the overall approach. On the other hand, in the chaos perturbing phase, a brand-new set of rules is presented to perturb the velocities and positions of particles to satisfy the ideal global search capability and adaptability, effectively avoiding the premature convergence problem found frequently in traditional PSO algorithm. In the above two stages, we control the number of selected items in each category to ensure the diversity of the final combination scheme. The fitness function of CS-PSO introduces the concept of the personalized constraints and general constrains to get a personalized interface, which is used to solve a personalized combination optimization problem. As part of our evaluation, we define a personalized dietary recommendation system, called Friend, where CS-PSO is applied to address a healthy diet combination optimization problem. Based on Friend, we implemented a series of experiments to test the performance of CS-PSO. The experimental results show that, compared with the typical HLR-PSO, CS-PSO can recommend dietary schemes more efficiently, while obtaining the global optimum with fewer iterations, and have the better global ergodicity.

The multiple-travelling salesman problem (MTSP) is a computationally complex combinatorial optimisation problem, with several theoretical and real-world applications. However, many state-of-the-art heuristic approaches intended to specifically solve MTSP, do not obtain satisfactory solutions when considering an optimised workload balance. In this article, we propose a method specifically addressing workload balance, whilst minimising the overall travelling salesman’s distance. More specifically, we introduce the two phase heuristic algorithm (TPHA) for MTSP, which includes an improved version of the K -means algorithm by grouping the visited cities based on their locations based on specific capacity constraints. Secondly, a route planning algorithm is designed to assess the ideal route for each above sets. This is achieved via the genetic algorithm (GA), combined with the roulette wheel method with the elitist strategy in the design of the selection process. As part of the validation process, a mobile guide system for tourists based on the Baidu electronic map is discussed. In particular, the evaluation results demonstrate that TPHA achieves a better workload balance whilst minimising of the overall travelling distance, as well as a better performance in solving MTSP compared to the route planning algorithm solely based on GA.

Through the availability of location-acquisition devices, huge volumes of spatio-temporal data recording the movement of people is provided. Discovery of the group of people who travel together can provide valuable knowledge to a variety of critical applications. Existing studies on this topic mainly focus on the movement of vehicles or animals with forcing the group members to stay always connected. However, the movement of people is different; people might belong to the same main group while they contribute in various sub-groups during their movement. In this paper, we propose a group pattern called loose travelling companion pattern (LTCP), which allows the members of a group to contribute to various sub-groups as long as the community of members does not change during the movement and all of the members stay connected for a few time-slots. In addition, we propose weakly continuous loose travelling companion pattern (WCLTCP) to relax the continuous time constraint in LTCP. Finally, three algorithms have been developed to discover the proposed group patterns: (i) straightforward approach, (ii) smart-and-fast method, and (iii) and opportunistic algorithm. Through the extensive experimental evaluation on both real and experimental datasets, the efficiency and effectiveness of the proposed group discovery approaches are proven.

We consider a special workflow scheduling problem in a hybrid-cloud-based workflow management system in which tasks are linearly dependent, compute-intensive, stochastic, deadline-constrained and executed on elastic and distributed cloud resources. This kind of problems closely resemble many real-time and workflow-based applications. Three optimization objectives are explored: number, usage time and utilization of rented VMs. An iterated heuristic framework is presented to schedule jobs event by event which mainly consists of job collecting and event scheduling. Two job collecting strategies are proposed and two timetabling methods are developed. The proposed methods are calibrated through detailed designs of experiments and sound statistical techniques. With the calibrated components and parameters, the proposed algorithm is compared to existing methods for related problems. Experimental results show that the proposal is robust and effective for the problems under study.

Indoor localization service based on magnetic field attracts increasing attention in the field of mobile computing in recent years. However, most of existing indoor localization methods based on magnetic field usually require the pedestrians to be equipped with a unified localization device, or take the magnetic value as the only characteristic of an indoor fingerprint map. Neither of these two kinds of methods could make sure that the indoor localization is ubiquitous, accurate and robust. Therefore, we propose a novel indoor hybrid fingerprint map and a robust Extended Particle Swarm Optimization Algorithm (EPSO). The proposed indoor hybrid fingerprint map is characterized by the variation trend of magnetic value and visible light intensity, which is suitable for most current smartphones that are usually integrated with the electronic compass and the light sensor. EPSO can correct not only the "data drifting" problem of sensors but also the logic error of localization called "wall crash". In addition, EPSO is still able to provide a high and robust localization accuracy when the attitude of smartphone is changed. Finally, we conduct a series of experiments and the experimental results show that EPSO is robust and it can achieve an accuracy within 1.7 m in the case of 80% localization errors.

Most existing indoor localization algorithms based on Wi-Fi signals mainly rely on wireless access points (APs), i.e. hotspots, with fixed deployment, which are easily affected by the non-line of sight (NLOS) factors and the multipath effect. There also exist many other problems, such as positioning stability and blind spots, which can cause decline in positioning accuracy at certain positions, or even failure of positioning. However, it will increase the hardware cost by adding more static APs; if the localization mechanism integrates different wireless signals is adopted, it tends to cause high cost of positioning and long complex positioning process, etc. In this paper, we proposed a novel hybrid Wi-Fi access point-based localization algorithm (HAPLA), which utilizes the received signal strength indications (RSSI) from static APs and dynamic APs to determine location scenes. It flexibly selects available AP signals and dynamically switches the positioning methods, thus to achieve efficient positioning. HAPLA only relies on the Wi-Fi signal strength values, which can reduce the cost of hardware and the complexity of localization system. The proposed method can also be able to effectively prevent interference from different signal sources. In our test scenario, we deployed typical indoor scenes with the NLOS factors and the multipath effect for experiments. The experiments demonstrate the effectiveness of proposed method and the results show that, compared with the classic K nearest neighbor-based location algorithm (KNN) and the variance-based fingerprint distance adjustment algorithm (VFDA), HAPLA has better adaptability and higher positioning accuracy, and can effectively solve the problem of positioning blind spots.

In this paper, we consider the widespread multi-stage job scheduling problem (e.g., in big data processed by MapReduce) in which jobs arrive at hybrid cloud systems stochastically. The objective is to minimize the number of elastic computing instances. Along with hard deadlines of jobs, the problem under study is NP-hard in strong sense. In terms of initial job priorities, timetables are constructed by adjusting job priorities adaptively and generating feasible schedules iteratively. Job sequences are generated by two simple dispatching rules. A fast local search heuristic and a rescheduling process are developed for improving the obtained sequences. Experimental results show that the proposed heuristics improve the utilization of computing resources effectively while meeting the cloud service quality requirements.