Achieving full-view coverage of dynamic targets in camera sensor networks (CSNs) remains a significant challenge, particularly in large-scale and dynamic environments where traditional optimization-based methods struggle to achieve efficient coordination. This study aims to develop an adaptive and scalable framework that enables coordinated sensing and dynamic reconfiguration of camera agents to maximize full-view coverage in real time. To this end, we propose an Automatic Grouping Algorithm (AGA) that reformulates the full-view coverage problem as a multi-agent cooperativecompetitive learning process. AGA integrates three synergistic mechanisms: (i) the Inter-group Competition Mechanism (IeCmM), which enhances adaptive camera positioning through competitive group evolution; (ii) the Intra-group Collaboration Mechanism (IrClM), which optimizes local sensing parameters to improve the coverage rate; and (iii) the Adaptive Adjustment Mechanism (AAM), which dynamically reconfigures group assignments to maintain optimal coverage under environmental variations. These modules are jointly optimized under a multi-agent reinforcement learning (MARL) framework with coordinated policy updates. Extensive experiments demonstrate that AGA consistently outperforms both traditional and state-of-the-art MARL-based baselines, achieving 16%–44% higher full-view coverage efficiency than traditional methods and beyond 71% improvement over advanced learning-based approaches such as HYGMA, HGAP, and GACG. Moreover, AGA exhibits rapid convergence, reaching near-optimal coverage within 10 episodes, and strong scalability across dynamic large-scale environments, demonstrating its effectiveness in complex CSN deployments.

This paper proposes a multi-objective optimization method for mixed-precision computation. Unlike previous studies that often take mantissa length reduction as the only optimization target, our work models the actual performance and power consumption of mixed precision programs on the corresponding hardware platforms, and based on this model searches for the pareto optimal set of all precision configurations. Experiments show that this tool can obtain performance improvements of 15% - 71 % on floating-point benchmarks while satisfying accuracy requirements. Compared to some typical counterpart-work, an average 21 % improvement can be obtained in SIMD scenarios.

As a legacy from conventional wireless services, illegal eavesdropping is regarded as one of the critical security challenges in Connected and Autonomous Vehicles (CAVs) network. Our work considers the use of Distributed Kalman Filtering (DKF) and Deep Reinforcement Learning (DRL) techniques to improve anti-eavesdropping communication capacity and mitigate jamming interference. Aiming to improve the security performance against smart eavesdropper and jammer, we first develop a DKF algorithm that is capable of tracking the attacker more accurately by sharing state estimates among adjacent nodes. Then, a design problem for controlling transmission power and selecting communication channel is established while ensuring communication quality requirements of the authorized vehicular user. Since the eavesdropping and jamming model is uncertain and dynamic, a hierarchical Deep Q-Network (DQN)-based architecture is developed to design the anti-eavesdropping power control and possibly channel selection policy. Specifically, the optimal power control scheme without prior information of the eavesdropping behavior can be quickly achieved first. Based on the system secrecy rate assessment, the channel selection process is then performed when necessary. Simulation results confirm that our jamming and eavesdropping defense technique enhances the secrecy rate as well as achievable communication rate compared with currently available techniques.

Vehicle-to-vehicle (V2V) communication applications face significant challenges to security and privacy since all types of possible breaches are common in connected and autonomous vehicles (CAVs) networks. As an inheritance from conventional wireless services, potential eavesdropping is one of the main threats to V2V communications. In our work, the anti-eavesdropping scheme in CAVs networks is developed through the use of cognitive risk control (CRC)-based vehicular joint radar-communication (JRC) system. In particular, the supplement of off-board measurements acquired using V2V links to the perceptual information has presented the potential to enhance the traffic target positioning precision. Then, transmission power control is performed utilizing reinforcement learning, the result of which is determined by a task switcher. Based on the threat evaluation, a multiple armed bandit problem is designed to implement the secret key switching procedure when it is needed. Through constant perception-execution loops (PELs), the security and confidentiality is improved for the authorized vehicles in their behavioral interactions with the illegal eavesdropper. Numerical experiments have presented that the developed approach has anticipated performance in terms of some risk assessment indicators.

Vehicle-to-vehicle (V2V) communication applications face significant challenges to security and privacy since all types of possible breaches are common in connected and autonomous vehicles (CAVs) networks. As an inheritance from conventional wireless services, illegal eavesdropping is one of the main threats to Vehicle-to-vehicle (V2V) communications. In our work, the anti-eavesdropping scheme in CAVs networks is developed through the use of cognitive risk control (CRC)-based vehicular joint radar-communication (JRC) system. In particular, the supplement of off-board measurements acquired using V2V links to the perceptual information has presented the potential to enhance the traffic target positioning precision. Then, transmission power control is performed utilizing reinforcement learning, the result of which is determined by a task switcher. Based on the threat evaluation, a multi-armed bandit (MAB) problem is designed to implement the secret key selection procedure when it is needed. Numerical experiments have presented that the developed approach has anticipated performance in terms of some risk assessment indicators.