Attention tracking in education can serve as a valuable tool for both students and teachers, particularly in the context of online learning. The objectives of our study were to establish the feasibility of developing a non-medical and low-cost dry-EEG neurofeedback system for the general public, usable in everyday life, out of the laboratory, with no need for further expert neurophysiological assistance, in order primarily to monitor and secondarily to optimize the attention levels of university students while watching e-learning videos. A neuromarker for attention was successfully determined using wearable electroencephalography (EEG) in real-life conditions. Assessing the pedagogical benefits of two types of multimodal neurofeedback, either personalized (for each student based on their individual attention level) or aggregate (identical for all students based on a previous cohort’s average attention level), we demonstrated that personalized neurofeedback significantly increased students’ attention levels following an attention drop. On the contrary, the aggregate neurofeedback yielded no positive impact on attention and was perceived as disruptive. This study underlines the feasibility and benefits of providing personalized feedback tailored to individual students during online learning in everyday life conditions, outside of the laboratory.

Assuring the trustworthiness of AI-based systems, especially in video surveillance, is critical in smart cities today. Large Language Models have emerged as the strongest tool for analyzing videos by offering human-readable descriptions of complex scenes. However, their effectiveness in sensitive tasks like abnormal behavior detection depends heavily on trust and system reliability. This paper proposes a new approach for enhancing trustworthiness in LLMs using a QoE-based framework. We propose a system that couples these Vid-LLMs with a real-time feedback loop, which incorporates both user corrections and validations to improve accuracy and user trust. We show how such a system can be trained and evaluated using the UCA Crime database and demonstrate that it is able to describe abnormal events in a highly reliable way. Our results indicate that the QoE-based trust model significantly improves user satisfaction, serving as a valued approach in real-world surveillance applications or other similar use cases.

The emergence of 6G technology is set to revolutionize connected autonomous vehicles (CAVs) by enabling hyper-connectivity, ultra-reliable low-latency communication, and seamless integration with IoT systems. These advancements enhance CAV efficiency, intelligence, and real-time data exchange for safer navigation and decision-making. However, the interconnected nature of 6G-enabled CAVs introduces significant cybersecurity risks, including adversarial AI attacks and large-scale intrusions. Traditional security methods are inadequate for addressing the complexity and sophistication of emerging threats in this dynamic ecosystem. This article aims to address these critical challenges by proposing an innovative security framework tailored for 6G-enabled CAVs. By integrating blockchain technology, reinforcement learning, and collaborative intrusion detection systems, this framework aims to secure CAV communications against malicious intrusions and ensure the reliability of their AI-driven operations. The system was evaluated on a dataset containing 2D image representations of normal and malicious network traffic. The ensemble-based IDS demonstrated high detection accuracy (99%) with low false positive rates. The Q-learning agent effectively supported trust-based consensus by reliably selecting validators and isolating malicious nodes. The blockchain layer maintained stable validation and propagation times, confirming the framework’s scalability and low-latency performance.

Achieving good-quality video frames and minimizing transmission delays are some of the most challenging requirements in the telemedicine system. The transmission process for real-time video streaming over wireless networks is affected by various real-time constraints, such as encoding mechanisms, noise, and bandwidth fluctuations, which can adversely affect the reliability and quality of the video transmission system. This work proposes a cross-layer system designed to enhance the Quality of Experience (QoE) and Quality of Service (QoS) for streaming video in low-latency applications over mobile health networks (m-Health), employing a Region-Of-Interest (ROI) video coding-based solution. Our system employs a modified extension of the multipath QUIC (MPQUIC), tailored to the context of emerging networks based on a 5G Non-Terrestrial Network (NTN), facilitating the reliable transmission of priority data through cellular networks and the unreliable transmission of non-priority data via a satellite link. Given the significance of the ROI, our system prioritizes these regions for reliable streaming, ensuring high-quality reception while sacrificing background quality by employing unreliable streaming. Simulation results demonstrate that our approach improves upon both state-of-the-art MPQUIC protocols and unreliable QUIC protocols. Specifically, the overall video quality increases by up to 40%, with up to a 50% improvement in ROI region quality compared to unreliable QUIC. Additionally, end-to-end delay is reduced by up to 35% compared to MPQUIC delays. Evaluations of both QoS and QoE have been carried out to validate the proposed approach. Furthermore, Our modified MPQUIC records a significant improvement in data rate measurements by up to 86% compared to the classical MPQUIC.

The advent of 6G technology has paved the way for unprecedented advancements in the Internet of Things (IoT), ushering in an era of hyper-connectivity and ubiquitous communication. However, with the proliferation of interconnected devices in 6G-enabled IoT ecosystems, the risk of malicious intrusions and new cyber threats becomes more prominent. Furthermore, the incorporation of AI into 6G networks introduces additional security concerns, such as the risk of adversarial attacks on AI models and the potential misuse of AI for cyber threats. Consequently, securing the extensive and diverse array of connected devices poses a substantial challenge in the 6G environment and needs reconsideration of prior security traditional methods. This paper aims to address these challenges by proposing a novel collaborative intrusion detection system (CIDS) that relies on AI and blockchain technologies. The collaborative nature of the proposed CIDS fosters a collective defense approach, where nodes within the IoT network actively share threat intelligence, enabling rapid response and mitigation. The effectiveness of the proposed system is evaluated through comprehensive simulations and proof-of-concept experiments. The results demonstrate the system’s ability to effectively detect and mitigate falsified and zero-day attacks, thereby fortifying the security infrastructure of 6G -enabled IoT environments.

In the multimedia field, the quality of experience (QoE) is rightfully seen as the center metric in research, around which each piece of the network is designed, especially for next-generation networks. Even if an increasing number of models using various data as inputs are now available, some major problems remain, such as the implementation of quality of experience measurement for real-time applications. The 'Human Factors' are the primary reason for the impossibility to correctly predict the QoE for real-time applications since these factors can't be measured easily and swiftly. For this reason, we present in this paper a Multi-Task model to predict multiple QoE influence factors at once from physiological data to save time in the training process and during the prediction. To test the model, we use a publicly available dataset named SoPMD, which contains recordings from an electroencephalogram (EEG), an electrocardiogram (ECG), and respiratory signals obtained during a quality assessment experiment, where QoE factors are gathered as labels. Our Multi-Task model has been tested using different features extracted from EEG and presents results up to 68.51% in accuracy. This model could be used in a real-time regulation loop to predict QoE factors faster than single-task models, for an enhanced QoE prediction, as this model can predict the five factors at the same time.

Multipath scheduling has been a hot research topic in recent years thanks to the development of multipath transport protocols (MP-TCP and MP-QUIC) and the wide deployment of 5G infrastructures. These schedulers take advantage of multiple physical interfaces on end-user devices, thus improving performance and reliability. While Peekaboo, the state-of-the-art scheduler, seems to provide the best performance in some specific environments, some evidence is that other existing schedulers may outperform it in more realistic network conditions. Based on this observation, we proposed GADaM, stands for Generic Adaptive Deep-learning based Multipath scheduler selector. GADaM’s role is to select the most appropriate scheduler under specific network conditions dynamically. The proposed paradigm proved its effectiveness in simulations, but its behavior in more realistic (typically mobile) environments remains largely unknown. This paper presents an experimental evaluation of GaDAM in real-world scenarios performed in an urban area near Paris, France. We design and implement a framework to serve this purpose, which takes mobility and fairness into consideration. Our results confirm the advantage of GaDAM’s approach compared to the deterministic scheduler selection approach.

Multipath QUIC (MQ-QUIC) and Multipath TCP (MP-TCP), known as multipath protocols, introduced several certain advantages for the next internet generation, such as enabling bandwidth aggregation of links, preventing single-path failure, increasing Quality of Service (QoS), etc. Meanwhile, the pivotal point of the transport protocols is the scheduler. Various multipath schedulers have been proposed, and each of them usually outperforms the others in each specific scenario. To provide a generic approach with the best performance and stability, a novel one is introduced in this paper and aimed to fill this research gap. Indeed, the proposed GADaM prototype is a Generic Adaptive Deep-learning-based Multipath Scheduler Selector. The idea’s prototype is implemented for the MP-QUIC protocol. The extensive results show that our scheduler selector achieved over 95% accuracy on training and 91% accuracy on the testing set in the simulated environment.