Intelligent Transportation Systems (ITS) is pivotal to smart city development, relying on Internet of Things (IoT) sensors deployed along roadways. In ITS, traffic flow prediction plays a crucial role. However, the complex spatio-temporal patterns in traffic data lead to information redundancy, which increases model complexity and limits prediction performance. Existing prediction methods, ranging from statistical models to deep learning approaches, typically process such correlated spatial and temporal features indiscriminately, resulting in and limited performance gains. To address this issue, this paper proposes a novel Coupled Adaptive Sparse Graph Convolutional Network (CASGCN) for traffic prediction. The core of the method is a coupled adaptive sparse strategy that simultaneously sparsifies the input features (node features and edge features) and GCN weights to improve the learning efficiency. Specifically, we design a bi-view adaptive sparse input features module that employs spatio-temporal projection and edge construction methods to selectively preserve informative features and essential connections. Meanwhile, we propose an adaptive sparse weighting scheme for the GCN, which incorporates orthogonal constraints to promote independence across convolutional filters and effectively mitigate redundancy. Extensive experiments on four real-world traffic datasets demonstrate that CASGCN achieves competitive performance compared to state-of-the-art methods, validating the effectiveness of our framework in improving traffic prediction.

Real-world IoT systems often require collaborative learning across multiple tasks performed by heterogeneous devices with varying computational resources. Conventional multi-task federated learning typically shares a pre-trained encoder across clients and trains task-specific classifiers locally. However, freezing the encoder during training leads to over-reliance on pre-training data and limits the model’s ability to capture inter-task relationships, reducing multi-task learning accuracy. Furthermore, significant variations in device capabilities highlight the need for heterogeneous model adaptation, making the unified model architecture assumption impractical for real-world IoT scenarios. To address these challenges, we propose MTFedSD, a novel multi-task federated learning across heterogeneous systems via self-distillation method. MTFedSD introduces multiple models of varying sizes and architectures to match clients’ resource capabilities, while enabling selective parameter sharing to extract consistent and transferable representations across tasks. A heterogeneous model aggregation strategy and self-distillation mechanism are integrated to reconcile differences across tasks and models, mitigating conflicts during joint training. Furthermore, the joint optimization strategy of shared encoder and pruning improves the communication efficiency in multi-task scenarios. Moreover, we provide a theoretical analysis that characterizes the convergence behavior of MTFedSD under heterogeneous and multi-task settings. Extensive experiments show that MTFedSD achieves robust and scalable performance across diverse tasks and device types, offering a practical solution for federated learning in complex IoT environments.

Federated learning (FL), an emerging distributed learning paradigm, addresses challenges in decentralized environments, particularly in internet of things (IoT) networks where numerous devices generate vast amounts of private data. A major concern in FL for IoT systems is communication overhead, especially in resource-constrained and wireless environments where frequent model uploads for aggregation create a critical bottleneck. As the complexity of neural networks increases, traditional FL methods demand substantial communication resources, which limits scalability in IoT applications. To address this, we propose a novel sparse communication mechanism for FL, called FedSC, that achieves high generalization performance and accuracy under extremely low communication frequencies, particularly for non-IID data. On the client side, we propose a multi-level model compression mechanism to capture and retain important information from the trained local model, which is then uploaded to the server. On the server side, we propose an iterative extraction mechanism to reconstruct models of uniform size based on the client models. Each extraction is followed by an aggregation step in an iterative process, ensuring effective generalization with non-IID data. As a result, clients need to upload their models only 1-5 times after completing local training, significantly reducing the communication overhead compared to traditional FL, which requires continuous uploads throughout the entire training process. Simulations on public datasets with popular deep learning models demonstrate that FedSC reduces the number of model uploads from hundreds to just 1-5 times while maintaining high accuracy, highlighting its potential to significantly enhance communication efficiency for FL in IoT systems.

Federated learning (FL), as an emerging collaborative learning paradigm, offers a promising solution to tackle learning challenges in IoT environments. However, the inherent mechanism of FL often results in significant communication overhead, which poses challenges in resource-constrained IoT systems. Moreover, most existing model compression methods struggle to balance model accuracy and compression ratio effectively with low resource consumption: excessive compression degrades accuracy, while insufficient compression incurs high communication overhead. To address this issue, we propose adaMC, an adaptive model compression algorithm for FL in IoT systems. This method integrates two types of network models: main and auxiliary networks. The auxiliary networks generate optimal sparsification strategies to guide the corresponding layers in the main network to become sparse. To ensure robust performance, we develop a bias correction method that guarantees both model accuracy and convergence, accompanied by a theoretical analysis. Experimental results on widely used deep learning models and public datasets demonstrate that our adaMC method is effective, achieving competitive performance compared to FL without compression.

It is necessary to establish a spatio-temporal correlation model in the traffic data to predict the state of the transportation system. Existing research has focused on traditional graph neural networks, which use predefined graphs and have shared parameters. But intuitive predefined graphs introduce biases into prediction tasks and the fine-grained spatio-temporal information can not be obtained by the parameter sharing model. In this paper, we consider it is crucial to learn node-specific parameters and adaptive graphs with complete edge information. To show this, we design a model based on graph structure that decouples nodes and edges into two modules. Each module extracts temporal and spatial features simultaneously. The adaptive node optimization module is used to learn the specific parameter patterns of all nodes, and the adaptive edge optimization module aims to mine the interdependencies among different nodes. Then we propose a Decoupled Adaptive Graph Convolution Attention Network for Traffic Forecasting (DAGCAN), which relies on the above two modules to dynamically capture the fine-grained spatio-temporal relationships in traffic data. Experimental results on four public transportation datasets, demonstrate that our model can further improve the accuracy of traffic prediction.

The increasingly popular federated learning still faces the practical challenge of non-independent and identically distributed data. Most efforts to address this issue focus on limiting local updates or enhancing model aggregation. However, these methods either restrict the learning capacity of local models or overlook the negative knowledge transfer caused by local objective divergences. In contrast, we observe that the global update can be re-expressed as a weighted sum of the gradients of samples from different classes. Therefore, we hypothesize that the competition among local updates may arise from the conflict between the gradients of samples belonging to different classes. Inspired by this insight, we introduce the novel perspective of orthogonal class gradients, aimed at eliminating interference between updates from different classes without the aforementioned drawbacks. To this end, this paper presentsFedOCF, which implements orthogonal class gradient constraints by encouraging orthogonality among features of different classes. Specifically,FedOCFmaintains a generator to learn features that are orthogonal for different classes and utilizes it to regularize features learned in local learning. Theoretically, we also demonstrate thatFedOCFcan improve generalization performance through feature conditional distribution alignment during local learning. Extensive experiments validate the excellent performance ofFedOCFin various heterogeneous scenarios.

Nowadays, most edge devices running Android still adopt the memory reclaim scheme designed for the server, which easily leads to a large number of page re-faults in the kernel, making mobile device workloads inefficient. We hold the opinion that the amount of reclaimable pages is sequential and predictable. Therefore, we propose an adaptive lightweight memory management scheme based on prediction, which consists of two parts: a lightweight and optimal prediction model generation module (LOPMG) and an adaptive prediction-based reclaim scheme (APRS). LOPMG adopts the Particle Swarm Optimization (PSO) algorithm to generate and quantify Long-Short Term Memory (LSTM) prediction model. By APRS, the PSO-LSTM model works in the kernel and can predict allocation workloads to tune reclaim parameters dynamically. Experiments show that our PSO-LSTM model can improve the accuracy of prediction, and the reclaim scheme can significantly reduce the number of page re-faults and the time of application relaunch.

Federated Learning (FL) has attracted extraordinary attention from the industry and academia due to its advantages in privacy protection and collaboratively training on isolated datasets. Since machine learning algorithms usually try to find an optimal hypothesis to fit the training data, attackers also can exploit the shared models and reversely analyze users’ private information. However, there is still no good solution to solve the privacy-accuracy trade-off, by making information leakage more difficult and meanwhile can guarantee the convergence of learning. In this work, we propose a Loss Differential Strategy (LDS) for parameter replacement in FL. The key idea of our strategy is to maintain the performance of the Private Model to be preserved through parameter replacement with multi-user participation, while the efficiency of privacy attacks on the model can be significantly reduced. To evaluate the proposed method, we have conducted comprehensive experiments on four typical machine learning datasets to defend against membership inference attack. For example, the accuracy on MNIST is near 99%, while it can reduce the accuracy of attack by 10.1% compared with FedAvg. Compared with other traditional privacy protection mechanisms, our method also outperforms them in terms of accuracy and privacy preserving.

Connected and autonomous vehicles offer the possibility to carry out control strategies, thus having great potential to improve traffic efficiency and road safety. The efficient passage of an emergency vehicle calls for the collaborative driving decision-making among multiple vehicles in a dynamically changing local area. However, existing work fails to efficiently adapt to dynamic and complex traffic conditions, thus cannot well solve the task. For better solution, we propose a flexible cooperative multi-agent reinforcement learning approach based on value function factorization, called Q-LSTM. Since the traffic environment is partially observable, the centralized training and decentralized execution paradigm is adopted to learn effective cooperative strategies for individual agents. To flexibly adapt to the changing neighborhood condition around the emergency vehicle, we introduce a long short-term memory network to decompose the learned global value function into local value function of each agent within the neighborhood, whose quantity and entities vary over time. To address the credit assignment problem and realize different roles of the emergency and regular vehicles, reward mechanism and the way agent-wise Q-networks update are well-designed. Extensive experiments are conducted on the Simulation of Urban MObility platform. Results show that our Q-LSTM outperforms state-of-the-art value-based MARL methods. Moreover, the robustness and adaptability of the Q-LSTM are verified in the cases of increased traffic density.

Federated learning (FL) is a promising distributed machine learning scheme where multiple clients collaborate by sharing a common learning model while maintaining their private data locally. It can be applied to a lot of applications, e.g., training an automatic driving system by the perception of multiple vehicles. However, some clients may join the training system dynamically, which affects the stability and accuracy of the learning system a lot. Meanwhile, data heterogeneity in the FL system exacerbates the above problem further due to imbalanced data distribution. To solve the above problems, we propose a novel FL framework named FedREM (Retain-Expansion and Matching), which guides clients training models by two mechanisms. They are 1) a Retain-Expansion mechanism that can let clients perform local training and extract data characteristics automatically during the training; 2) a Matching mechanism that can ensure new clients quickly adapt to the global model based on matching their data characteristics and adjusting the model accordingly. Results of extensive experiments verify that our FedREM outperforms various baselines in terms of model accuracy, communication efficiency, and system robustness.

Traffic forecasting, which benefits from mobile Internet development and position technologies, plays a critical role in Intelligent Transportation Systems. It helps to implement rich and varied transportation applications and bring convenient transportation services to people based on collected traffic data. Most existing methods usually leverage graph-based deep learning networks to model the complex road network for traffic forecasting shallowly. Despite their effectiveness, these methods are generally limited in fully capturing high-order spatial dependencies caused by road network topology and high-order temporal dependencies caused by traffic dynamics. To tackle the above issues, we focus on the essence of traffic system and propose STHODE: Spatio-Temporal Hypergraph Neural Ordinary Differential Equation Network, which combines road network topology and traffic dynamics to capture high-order spatio-temporal dependencies in traffic data. Technically, STHODE consists of a spatial module and a temporal module. On the one hand, we construct a spatial hypergraph and leverage an adaptive MixHop hypergraph ODE network to capture high-order spatial dependencies. On the other hand, we utilize a temporal hypergraph and employ a hyperedge evolving ODE network to capture high-order temporal dependencies. Finally, we aggregate the outputs of stacked STHODE layers to mutually enhance the prediction performance. Extensive experiments conducted on four real-world traffic datasets demonstrate the superior performance of our proposed model compared to various baselines.

Situational awareness tries to grasp the important events and circumstances in the physical world through sensing, communication, and reasoning. Tracking the evolution of changing situations is an essential part of this awareness and is crucial for providing appropriate resources and help during disasters. Social media, particularly Twitter, is playing an increasing role in this process in recent years. However, extracting intelligence from the available data involves several challenges, including (a) filtering out large amounts of irrelevant data, (b) fusion of heterogeneous data generated by the social media and other sources, and (c) working with partially geo-tagged social media data in order to deduce the needs of the affected people. Spatio-temporal analysis of the data plays a key role in understanding the situation, but is available only sparsely because only a small fraction of people post relevant text and of those very few enable location tracking. In this paper, we provide a comprehensive survey on data analytics to assess situational awareness from social media big data.

Distributed machine learning (ML) was originally introduced to solve a complex ML problem in a parallel way for more efficient usage of computation resources. In recent years, such learning has been extended to satisfy other objectives, namely, performing learning in situ on the training data at multiple locations and keeping the training datasets private while still allowing sharing of the model. However, these objectives have led to considerable research on the vulnerabilities of distributed learning both in terms of privacy concerns of the training data and the robustness of the learned overall model due to bad or maliciously crafted training data. This article provides a comprehensive survey of various privacy, security, and robustness issues in distributed ML.

Missing data is a common and critical problem in the stage of traffic data collection and processing. How to impute the missing values in spatio-temporal traffic data has been a challenging topic for a long time. Recently, a variety of methods have been proposed to impute the missing values. Among them, the tensor-based methods show higher competence in multi-dimensional traffic data imputation. However, the previous studies of tensor factorization rarely considered the joint imputation of multiple correlative data such as traffic speed and traffic volume. In this paper, a novel method called Multi-Task Neural Tensor Factorization (MTNTF) is proposed to learn the non-linear correlation patterns of traffic speed-volume, and then address the joint imputation of traffic speed and traffic volume. Extensive experiments on a real dataset show our MTNTF significantly outperforms the state-of-the-art methods in element-wise missing and fiber-wise missing cases. In addition, our method can impute the slice-wise missing values of traffic volume based on incomplete traffic speed.

Driven by the advanced technologies of vehicular communications and networking, the Internet of Vehicles (IoV) has become an emerging paradigm in smart world. However, privacy and security are still quite critical issues for the current IoV system because of various sensitive information and the centralized interaction architecture. To address these challenges, a decentralized architecture is proposed to develop a blockchain-supported IoV (BS-IoV) system. In the BS-IoV system, the Roadside Units (RSUs) are redesigned for Mobile Edge Computing (MEC). Except for information collection and communication, the RSUs also need to audit the data uploaded by vehicles, packing data as block transactions to guarantee high-quality data sharing. However, since block generating is critical resource-consuming, the distributed database will cost high computing power. Additionally, due to the dynamical variation environment of traffic system, the computing resource is quite difficult to be allocated. In this paper, to solve the above problems, we propose a Deep Reinforcement Learning (DRL) based algorithm for resource optimization in the BS-IoV system. Specifically, to maximize the satisfaction of the system and users, we formulate a resource optimization problem and exploit the DRL-based algorithm to determine the allocation scheme. The evaluation of the proposed learning scheme is performed in the SUMO with Flow, which is a professional simulation tool for traffic simulation with reinforcement learning functions interfaces. Evaluation results have demonstrated good effectiveness of the proposed scheme.

Spatial clustering of the events scattered over a geographical region has many important applications, including the assessment of needs of the people affected by a disaster. In this paper we consider spatial clustering of social media data (e.g., tweets) generated by smart phones in the disaster region. Our goal in this context is to find high density areas within the affected area with abundance of messages concerning specific needs that we call simply as “situations”. Unfortunately, a direct spatial clustering is not only unstable or unreliable in the presence of mobility or changing conditions but also fails to recognize the fact that the “situation” expressed by a tweet remains valid for some time beyond the time of its emission. We address this by associating a decay function with each information content and define an incremental spatial clustering algorithm (ISCA) based on the decay model. We study the performance of incremental clustering as a function of decay rate to provide insights into how it can be chosen appropriately for different situations.

Spatial big data analysis is very important in disaster scenarios to understand distribution patterns of situations, e.g., people's movements, people's requirements, resource shortage situations, and so on. In a general case, spatial big data is generated from distributed sensing devices and analyzed in a centralized way, e.g., a cloud center with high-performance computing resources. However, data transmission from sensing devices to cloud centers always takes a long time, especially in disaster scenarios with an unstable network. Fog computing is a promising technique to solve the above problem by offloading data processing tasks from the cloud to nearby computation devices. But data resolution also decreases after local processing in the fog nodes. It is necessary to investigate the optimal task distribution solutions to efficiently use computation resources in the fog layer. In this paper, we take the above research problem, and study fog-computing supported spatial big data processing. We analyze the process for spatial clustering, which is a typical category for spatial data analysis, and propose an architecture to integrate data processing into fog computing. We formalize a problem to maximize the data-resolution efficiency by considering data resolution and delay. We further propose core algorithms to enable spatial clustering in a fog-computing environment and implement the above algorithms in a real system. We have performed both simulations and experiments on a real Twitter dataset collected when Kumamoto-city suffered an earthquake. Through the simulations and the experiments, we have determined that the proposed solution significantly outperforms the other solutions.

Traditional robotic systems are often incapable of handling complex tasks due to hardware constraints, such as computing ability, storage space, and battery capacity. Cloud robotic systems, characterized by allowing multi-robot systems to access the powerful cloud infrastructures, is a promising solution to fulfill complex tasks, such as disaster management, real-time object recognition, 3D Simultaneous Localization And Mapping (SLAM). However, the destabilizing factors of network could lead to high latency of data transmission in cloud robotic systems, which have made great challenges to the fields that have high real-time requirements. What's more, the existence of heterogeneity of robots further complicates cloud robotics cooperation. In order to minimize the average response time in latency-aware scenarios, we jointly investigate task assignment and scheduling in Collaborative Cloud Robotic Systems (CCRS). We first formulate the problem into a Mixed-Integer Non-Linear Programming (MINLP) and then linearize it into an Integer Linear Programming (ILP) using discrete time structure. To meet the extensibility requirement, we further propose a partitioning-based algorithm to deal with large-scale task graphs. The results show that our two approaches outperform the existing genetic algorithm and greedy algorithm.

Big data analytics has started to use data collected from the sensors in smartphones. This data may be used by disaster response teams for locating problems. But the regular communication infrastructure can be destroyed after disasters. Movable base stations (MBS), as studied by the company NTT, offer an easily deployable solution to construct an emergency communication network~ (ECN), but are not suitable for transmitting big data from sensing devices to the cloud for data processing in the cloud. To address this issue, MBSs have been equipped with processing capabilities of their own, which creates an MBS-based Fog-computing Network. We proposed a novel algorithm to minimize the overall cost of the system while maintaining 0 data overflow. This will allow the resources to be used at the most efficient level. Our genetic algorithm solution had a reduced system cost over various network sizes when compared to some conventional solutions. During the simulation, it was clear that the best conventional method for preventing data overflow was the fog-based solution, but its cost was quite high. The cloud-based solution had the lowest cost but would lead to a large amount of data overflow, which would need to be cached. The GA-based solution maintained the ideal solution throughout the variation of all bandwidth parameters: the processing rate, the data compression ratio, and the cost coefficient ratio. Because none of the conventional solutions were able to match the capabilities of the GA for the current constraints, we believe that this should be investigated further with a faster algorithm.

Spatial big data analytics has become possible with the data collected from the sensors in smart phones, which can support decision-making in disaster scenarios. However, sometimes the regular communication infrastructure can be destroyed after disasters. Movable base stations (MBS), as studied by the company NTT, offer an easily deployable solution to construct an emergency communication network, but are not suitable for transmitting big data from sensing devices to the cloud for data processing in the cloud. To solve this issue, we studied a novel algorithm to process spatial big data efficiently in a wirelessly networked disaster area that uses multiple MBSs. More specifically, we proposed a novel algorithm to minimize overall delay for spatial data processing in wirelessly-networked disaster areas (SDP-WNDA), to enable quick responses to data analysis. Our proposed model and genetic algorithm solution showed to have a reduced maximum end- to-end (E2E) delay over various network sizes, when compared to some conventional solutions. For the realistic constraints, the cloud solution was the best conventional method, followed by the system which used the fog nodes to process as much data as possible, but the genetic algorithm (GA) had a slight advantage over all other methods. However, as the computation rate, μk, was increased, the maximum processing algorithm got much stronger. Also, as the communication capacity, R, was increased, the cloud computing solution was more successful. The fact that none of the conventional cases matched the capabilities of the GA for increased computation or increased transmission rates suggests the need for this to be investigated even further.

Control and Communication in the computing environment with diverse equipment could be clumsy, obtrusive, and frustrating even just for finding the right input device or getting familiar with the input interface. In this paper, we present Magic Ring (MR), a finger ring shape input device using inertial sensor to detect the subtle finger gestures and routine daily activities. As a self-contained, always-available, and hands-free input device, we believe that MR will enable diverse applications in the intelligent computing environment. In this demonstration, we will show a prototype design of MR and three proof-of-concept application systems: a remote controller to control the electrical appliance like TV, radio, and lamp using simple finger gestures; a natural communication tools to chat using the simplified sign languages; a daily activity tracker to record daily activities such as room cleaning, eating, cooking, writing with only one MR on the index finger.

With the progress of sensing, computing, communication technologies, ubiquitous computing becomes a hot topic in the research field, by which various kinds of services can be provided to satisfy users. Ubiquitous learning as one application of ubiquitous computing is playing a very important role in our daily life. Training is a kind of learning in ubiquitous environment. During the whole training process, the performance of a learner in a series of units may change with patterns: progress, plateau, and decline. Therefore, it is necessary to apply different support strategies to adapt to different states of the learner. In the previous work, a mutual adaptive support based on one learner model was proposed. However one learner model may sometime make the learner lose his/her motivation easily, since each learner does the training independently. There is a need to propose support for group training which is used to let learners do training and improve themselves together. But in the current models for group training, most of them focus on cooperative and competitive training but ignore balance feeling of connection and burden of learners. Feeling of connection is very important in group training since it can give learners continuing motivation by exchanging information. However it also may bring burden to learners due to frequency, volume and content of exchanged information. So an effective mechanism is needed to balance the feeling of burden and connection of multiple learners. To this end, we propose a new support method for a pair of learners to solve the problem. In this paper, firstly, we calculate how much a learner affects the other learner. Then we design an algorithm to judge the feeling of connection and burden from the other leaner. Finally, due to different feeling of connection and burden, various support strategies are designed to support learners. We have implemented our support method in a typing game and shown the method can balance the feeling of connection and burden.

We study data fusion in sensor networks using mobile agents (MAs),which are capable of saving energy of sensor nodes and performing advanced computation functions based on the requests of various applications. Research on MAs still remains unfledged in development of application-oriented data fusion, which is highly desired in wireless sensor networks (WSNs) deployed in recent days for environmental and disaster monitoring. In this paper, we propose a dynamic itinerary planning for MAs (DIPMA) to collect data from sensor networks with an application-oriented approach. In particular, the DIPMA algorithm is applied to the data collection for frost prediction which is a real-world application in agriculture using next- generation sensor networks. The performance of the DIPMA is evaluated by simulations and the experimental results show that the total execution time of MA can be reduced significantly with our approach while sound prediction accuracy is maintained.