The dynamic optimization of large-scale transportation networks presents significant challenges due to their complexity, stochasticity, and the need for real-time decision-making. In conventional methods, there is often a failure to employ fully the resources of big data present in cities, thus not being able to respond appropriately to fluctuations in traffic conditions. This paper introduces a novel enhanced big data-driven reinforcement learning (EBD-RL) algorithm for dynamic optimization of transportation networks, addressing these challenges by leveraging advanced machine learning techniques and heterogeneous data sources. We propose a hierarchical control framework that decomposes the global optimization problem into manageable sub-problems while maintaining network-wide coordination. The EBD-RL algorithm incorporates prioritized experience replay and adaptive exploration strategies to improve learning efficiency and stability in high-dimensional state spaces. Experiments on a realistic urban network show that our method is superior to six state-of-the-art methods. Results show that EBD-RL reduces total travel time by up to 30% compared to the best baseline under various traffic demands and connected autonomous vehicle penetration rates. Furthermore, the algorithm exhibits enhanced resilience to traffic incidents, achieving up to 33% faster network recovery time in severe disruption scenarios. These findings highlight the potential of big data-driven reinforcement learning approaches to significantly improve the efficiency, adaptability, and resilience of modern urban transportation systems.

With the rapid advancement of technologies such as B5G/6G and edge computing, network scenarios are becoming increasingly complex and diverse, leading to the emergence of slicing networks. Virtualizing applications into distinct categories and establishing corresponding network slices ensures performance to a certain extent. However, the challenges posed by the complex slicing environment demand more fine-grained routing control and higher costs to locate requested content or services, areas where current state-of-the-art methods fall short. To address these challenges, this work introduces a system framework that integrates the principles of Segment Routing over IPv6 (SRv6). An SRv6 optimization layer is created between the control and infrastructure layers to manage slices effectively and enhance routing control. Additionally, we propose a novel policy routing method based on zero-trust and Graph Convolutional Network (GCN) technology. This method transforms actions into policies that can be flexibly deployed on SRv6 nodes, segment by segment. These actions encompass both routing and security measures, allowing for dynamic and flexible deployment of policies on each segment to achieve the desired goals. This integration of segment routing and zero-trust principles simplifies implementation and enhances security. Comprehensive experiments were conducted to evaluate the proposed method. The results demonstrate significant improvements over state-of-the-art methods, including a higher service acceptance rate, better resource utilization, and reduced average latency and packet loss rate.

Sentiment analysis of social media platforms is crucial for extracting actionable insights from unstructured textual data. However, modern sentiment analysis models using deep learning lack explainability, acting as black box and limiting trust. This study focuses on improving the explainability of sentiment analysis models of social media platforms by leveraging explainable artificial intelligence (XAI). We propose a novel explainable sentiment analysis (XSA) framework incorporating intrinsic and posthoc XAI methods, i.e., local interpretable model-agnostic explanations (LIME) and counterfactual explanations. Specifically, to solve the problem of lack of local fidelity and stability in interpretations caused by the LIME random perturbation sampling method, a new model-independent interpretation method is proposed, which uses the isometric mapping virtual sample generation method based on manifold learning instead of LIMEs random perturbation sampling method to generate samples. Additionally, a generative link tree is presented to create counterfactual explanations that maintain strong data fidelity, which constructs counterfactual narratives by leveraging examples from the training data, employing a divide-and-conquer strategy combined with local greedy. Experiments conducted on social media datasets from Twitter, YouTube comments, Yelp, and Amazon demonstrate XSAs ability to provide local aspect-level explanations while maintaining sentiment analysis performance. Analyses reveal improved model explainability and enhanced user trust, demonstrating XAIs potential in sentiment analysis of social media platforms. The proposed XSA framework provides a valuable direction for developing transparent and trust-worthy sentiment analysis models for social media platforms.

Sentiment analysis, a challenging task in understanding human emotions expressed through diverse modalities, prompts the development of innovative solutions. Multimodal data often contains important complementary information. Effective fusion and extraction of multimodal data features are key issues in sentiment analysis. In this article, we introduce a novel sentiment analysis model that integrates multimodal multiscale features based on a fuzzy-deep neural network. First, we combine multimodal data, namely text, audio, and images, to extract intrinsic feature representations. Second, our model incorporates the fuzzy-deep neural network learning module, infused with fuzzy logic principles to enhance adaptability to the inherent vagueness in sentiment expressions. Furthermore, we integrate the dual attention mechanism that dynamically focuses on pivotal aspects within multimodal data, refining feature extraction for heightened context-awareness. Rigorous validation across three datasets, including the Multimodal Corpus of Sentiment Intensity dataset, the Multimodal Opinion Sentiment and Emotion Intensity dataset, and the Chinese Single and Multimodal Sentiment dataset, demonstrates the model's superior performance in capturing the intricacies of human emotions.

Electrocardiogram (ECG) analysis and diagnosis are important auxiliary means for preventing and detecting cardiovascular diseases. Traditional approaches often face challenges due to the sheer volume of data, difficulty in extracting meaningful features, limitations in model complexity, and the requirement for real-time analysis in clinical settings. This paper presents a pioneering approach for automatic ECG diagnosis through the application of a novel Multiscale Deep Neuro-fuzzy Network (MDNFN) structure. The MDNFN is designed to address the complexity of arrhythmia classification by incorporating deep learning and fuzzy logic processing across multiscale feature extraction. To optimize the performance of the MDNFN, an innovative model optimization technique based on the Particle Swarm Optimization (PSO) algorithm is introduced, offering an efficient exploration of the parameter space. Extensive experiments across diverse datasets validate the superior performance of the proposed model compared to existing methods. The MDNFN demonstrates heightened accuracy and robustness, supported by its adaptability to different frequency and time scales inherent in ECG signals. The study establishes the model's efficacy through comprehensive experimentation, providing compelling evidence for its potential application in real-world clinical scenarios.

Due to its superiority in addressing label ambiguity, label distribution learning (LDL) has received wide attention from the community, such as image classification, emotion recognition, and big data processing. To efficiently process the data with label distribution, researchers have proposed to learn label-specific features (LSFs) that are the discriminative features for each class label. Although the LDL literature has seen many algorithms to learn LSFs, most of them ignore the characteristics of label distribution. Label distribution lies in real-value vector space with specific characteristics. In this article, we propose to learn label-distribution-specific features (LDSFs) for processing label distribution data by considering the structures of label distribution. We design a novel LDL method called LDL-LDSF to exploit LDSFs by considering the fuzzy cluster structures of label distribution data. First, LDL-LDSF learns LDSFs for the whole label distribution by jointly learning the label distribution and fuzzy C-means clustering. Second, it learns LDSFs for each label in a similar way. Third, it concatenates the learned LDSFs with the original features to deduce an LDL model. Finally, we conduct extensive experiments to justify that LDL-LDSF statistically outperforms several state-of-the-art LDL methods and validate the advantages of LDSFs for processing label distribution data.

Ensuring the safety and reliability of Transportation Cyber-Physical Systems (T-CPS) is critical. However, the increasing interconnectedness of T-CPS exposes them to sophisticated cyberattacks, necessitating robust intrusion detection systems (IDS) to safeguard against evolving threats. This paper aims to enhance the security of T-CPS by addressing two key challenges: effective anomaly detection and handling imbalanced datasets in intrusion detection tasks. In this paper, we propose ReMeNet (Reconstruction Memory Network), a novel intrusion detection model that combines a memory module with a GAN-based architecture to enhance anomaly detection and data reconstruction. To address the challenge of imbalanced datasets, we incorporate a Vector Quantized Wasserstein Generative Adversarial Network (VQ-WGAN) to generate additional samples for underrepresented attack categories, thereby balancing the dataset and improving detection performance. Experimental evaluation on the UNSW-NB15 dataset demonstrates that ReMeNet achieves an accuracy of 91.70%, and an F1-score of 91.63% which outperforms Random Forest by 8.02% and EC-GAN by 3.01%. The results show that ReMeNet effectively handles imbalanced data, improving detection rates across all attack categories in T-CPS.

The integration of Artificial Intelligence with the Internet of Things significantly enhances the functionality of vehicle–road cooperation (VRC) systems by enabling smarter, real-time decision-making and resource optimization across interconnected vehicular networks. To tackle the challenges associated with resource constraints, this study introduces a method where vehicle users can offload tasks to nearby roadside units (RSUs) or service-oriented vehicles to ensure timely application execution. However, this task offloading introduces additional transmission delays and energy expenditures. Consequently, this article first conceptualizes the computation offloading problem, aiming to minimize the total task processing time and energy consumption under the constraints of resources provided by RSUs and service-oriented vehicles. We model the computation offloading issue within the VRC framework as a Markov decision process (MDP) and propose a multiagent reinforcement learning-based resource scheduling method. Each vehicle, acting as an intelligent agent, interacts with and influences decisions within this environment. The method integrates the twin delayed deep deterministic policy gradient algorithm to train deep neural networks for deciding on task offloading and computational resource allocation. Simulation results demonstrate that compared to existing algorithms, the proposed method more effectively utilizes the computational resources available through RSUs and service-oriented vehicles within the VRC system. It achieves joint optimization of latency and energy consumption, thus validating the efficacy of the proposed approach in enhancing the operational efficiency and sustainability of urban transportation systems.

Vehicular edge computing (VEC) systems face challenges in providing real-time intelligent transportation services due to limited computing resources at VEC servers, which lead to excessive delays or denial of services, especially for latency-critical tasks. This article proposes an augmented intelligence of things (AIoT) framework to enable priority-aware task offloading in VEC for vehicle road cooperation systems, maximizing overall system rewards under latency constraints. The framework incorporates an advanced dynamic resource management mechanism that adapts to real-time data and optimizes resource allocation using augmented intelligence models. The joint priority-aware application offloading and resource optimization problem is formulated as a constrained Markov decision process, and a deep Q-network (DQN)-based learning algorithm is employed to optimize the allocation of communication and computational resources based on application priorities and real-time channel/queue state information. Simulation results demonstrate that the proposed algorithm achieves significant improvements in weighted carrying capacity, high/low-priority task drop rates, and high/low-priority task queuing delays under varying overall task arrival rates, proportions of high/low-priority tasks, vehicle density, and task size compared to benchmark schemes. The proposed AIoT-enhanced DQN-based learning algorithm advances the field of VEC systems for vehicle road cooperation, offering practical advantages, such as increased efficiency, reduced latency, and improved resource utilization, ultimately enhancing user experience and enabling real-world applications in intelligent transportation systems.

Internet-connected robotic systems today predominantly rely on isolated sensing, computing, and communication modules, limiting cross-layer optimizations. However, emerging applications like industrial automation and augmented reality necessitate tight coupling between complementary capabilities for versatility, precision, and autonomy improvements. To this end, this paper proposes an adaptive sensing algorithm for Internet of Robotic Things (IoT) platforms with integrated sensing, computing, and communication (as-ISCC-IoRT) capabilities. The framework leverages a model-driven methodology to dynamically harness the benefits of complementary techniques for improving localization accuracy and operational efficiency. First, three classical sensing algorithms are introduced to realize multi-target ranging and speed measurement, and the algorithms are analyzed in terms of sensing accuracy, communication performance, and computational complexity, which shows that any one of the algorithms alone cannot achieve the optimization of sensing accuracy, sensing capacity, and communication rate simultaneously. Then, combining the characteristics of different sensing algorithms, an adaptive sensing algorithm is proposed, and the receiver selects the appropriate sensing algorithm based on the ratio of the measured received signal to the interference plus noise. Extensive simulations under varying signal-to-interference-plus-noise ratio levels, number of sensors, and quality of service constraints validate the effectiveness of as-ISCC-IoRT -consistently showing the fastest convergence, lowest weighted MSE, highest communications rate gain, and minimum transmit power by adaptively switching between component algorithms. The consistent performance gains of the proposed as-ISCC-IoRT scheme across key metrics like accuracy, latency, and efficiency validate the benefits of integrating sensing, computing, and communication capabilities in Internet-connected robotic systems.

The proliferation of multimedia-enabled IoT devices and edge computing enables a new class of data-intensive applications. However, analyzing the massive volumes of multimedia data presents significant privacy challenges. We propose a novel framework called generative adversarial privacy (GAP) that leverages generative adversarial networks (GANs) to synthesize privacy-preserving surrogate data for multimedia analytics across the IoT-Edge continuum. GAP carefully perturbs the GAN's training process to provide rigorous differential privacy guarantees without compromising utility. Moreover, we present optimization strategies, including dynamic privacy budget allocation, adaptive gradient clipping, and weight clustering to improve convergence and data quality under a constrained privacy budget. Theoretical analysis proves that GAP provides rigorous privacy protections while enabling high-fidelity analytics. Extensive experiments on real-world multimedia datasets demonstrate that GAP outperforms existing methods, producing high-quality synthetic data for privacy-preserving multimedia processing in diverse IoT-Edge applications.

The exponential growth of mobile data traffic in next generation networks has led to a significant increase in energy consumption, posing critical challenges for network operators. We propose DLLF-2EN, a novel energy-efficient framework that integrates deep learning-based load forecasting, an advanced power consumption model, and a comprehensive energy-saving strategy to address this issue. The load forecasting technique utilizes deep convolutional neural network and long short-term memory model, which is based on deep learning. This model is capable of capturing the spatiotemporal dependencies present in network traffic data. The power consumption model accurately characterizes the base stations’ static and dynamic power consumption components, facilitating the assessment of energy efficiency under various network scenarios. The energy-saving strategy combines base station sleep mode with discontinuous transmission and reception, as well as lightweight transmission of common signals, dynamically adapting the network operation based on the predicted traffic load. Furthermore, DLLF-2EN incorporates an intelligent power management system that leverages machine learning algorithms to continuously monitor the network, analyze collected data, and make optimal energy-saving decisions in real-time. Simulation demonstrate that the superior performance of DLLF-2EN in terms of load forecasting accuracy and energy efficiency compared to state-of-the-art baseline methods. The proposed framework represents a comprehensive solution for energy-efficient and sustainable next generation mobile networks, addressing the critical challenges of minimizing energy consumption while meeting the growing demands for high-quality mobile services.

Multi-cloud storage is recently a viable approach to solve the vendor lock-in, reliability, and security issues in cloud storage systems. As a key concern, data placement influences the cost and performance of storage services. Yet, in practice it remains challenging to address the huge solution space. Previous studies typically focus on constructing efficient data placement schemes based on the predicted pattern of workloads or assuming fully a-priori known network conditions. They cannot be easily applied in multi-cloud storage scenarios, which typically involve dynamic network conditions and time-varying workloads. To this end, we formulate the data placement optimization in a combinatorial multi-arm bandit (CMAB) perspective and solve it by learning placement strategy online. In contrast to a stationary setting where reward distributions are unknown but identical over time, we consider a realistic multi-cloud environment with non-stationary conditions, i.e., reward distributions change over time. To swiftly accommodate this, we propose an adaptive window combinatorial upper confidence bound based data placement (AW-CUCB-DP) scheme to reduce latency and cost. In AW-CUCB-DP, a simple and efficient change detector, i.e.,Page-Hinkley testwith forgetting mechanism (FM-PHT), is employed to enable variable-size sliding windows to handle both gradual and abrupt variations in network conditions or workloads. We establish that AW-CUCB-DP is asymptotically optimal in the non-stationary multi-cloud environment. Trace-driven experiments further verify that our scheme outperforms alternatives, especially in highly dynamic environments.