Approximate Nearest Neighbor Search (ANNS) is a cornerstone algorithm for information retrieval, recommendation systems, and machine learning applications. While x86-based architectures have historically dominated this domain, the increasing adoption of ARM-based servers in industry presents a critical need for ANNS solutions optimized on ARM architectures. A naive port of existing x86 ANNS algorithms to ARM platforms results in a substantial performance deficit, failing to leverage the unique capabilities of the underlying hardware. To address this challenge, we introduce KScaNN, a novel ANNS algorithm co-designed for the Kunpeng 920 ARM architecture. KScaNN embodies a holistic approach that synergizes sophisticated, data aware algorithmic refinements with carefully-designed hardware specific optimizations. Its core contributions include: 1) novel algorithmic techniques, including a hybrid intra-cluster search strategy and an improved PQ residual calculation method, which optimize the search process at a higher level; 2) an ML-driven adaptive search module that provides adaptive, per-query tuning of search parameters, eliminating the inefficiencies of static configurations; and 3) highly-optimized SIMD kernels for ARM that maximize hardware utilization for the critical distance computation workloads. The experimental results demonstrate that KScaNN not only closes the performance gap but establishes a new standard, achieving up to a 1.63x speedup over the fastest x86-based solution. This work provides a definitive blueprint for achieving leadership-class performance for vector search on modern ARM architectures and underscores

To meet the computational requirements of computation-intensive and delay-sensitive applications, we construct an Unmanned Aerial Vehicle (UAV)-assisted three-layer collaborative computing framework that integrates local, edge, and cloud computing resources. However, in dynamic UAV-assisted environments, some existing approaches lack adaptability and struggle to effectively balance delay and energy consumption. To address these challenges, we formulate a joint optimization problem that minimizes the weighted sum of delay and energy consumption, where adaptive weight factors are dynamically adjusted according to system state variations. Due to the non-convex and high-dimensional nature of our proposed problem, traditional optimization methods are generally inadequate. Hence, the problem is modeled as a Markov Decision Process (MDP), and a normalization-based reward function is designed to eliminate the dimensional imbalance between delay and energy consumption. A Dynamic Normalization Twin Delayed Deep Deterministic Policy Gradient (DN-TD3) algorithm is then proposed, which incorporates mechanisms of adaptive exploration and criticdriven policy updates to enhance convergence stability and reduce sensitivity to hyperparameters. Simulation results demonstrate that the proposed DN-TD3 algorithm outperforms benchmark schemes in terms of system cost reduction, convergence speed, and overall stability.

To enable compute-intensive and delay-sensitive maritime services, unmanned surface vessels (USVs) can offload tasks to mobile edge computing (MEC) servers mounted on unmanned aerial vehicles (UAVs). However, jointly minimizing energy consumption and latency is challenging due to the strong coupling between communication, computation, and mobility under stringent quality-of-service (QoS) requirements. To capture this trade-off, we formulate a weighted energy–delay minimization problem that jointly optimizes one-to-one UAV–USV scheduling, task partitioning, and UAV trajectory. The resulting problem is particularly difficult due to a hybrid discrete–continuous decision space and strong temporal coupling under stringent feasibility constraints. To address this mixed-integer nonconvex optimization problem, we reformulate it as a Markov decision process (MDP) and develop a constraint-aware OU–TD3 algorithm that integrates differentiable scheduling relaxation, feasibility-aware action mapping, and adaptive OU–Gaussian mixed exploration for stable learning in high-dimensional continuous control. We further extend the formulation and solution to a cooperative multi-UAV MEC setting with signal-to-interference-plus-noise ratio (SINR)-coupled interference and coordination constraints. Extensive simulations with statistical evaluation demonstrate stable convergence and up to 54.2% cost reduction over baseline schemes, while maintaining robustness under realistic maritime disturbances.

To improve the service quality for edge users, this paper proposes the full-duplex (FD) cooperative non-orthogonal multiple access (CNOMA) framework for a reconfigurable intelligent surface (RIS)-assisted integrated sensing and communication (ISAC) network. Specifically, OMA is integrated into CNOMA to mitigate substantial inter-cluster interference. Furthermore, we propose an optimization problem aimed at simultaneously maximizing the weighted sum rate and sensing power. This problem includes optimizing the power allocation of the near user in a cooperative system, the power allocation factors within each cluster, the beamforming matrix at the base station (BS) and the phase shift matrix at the RIS. To address this highly-coupled and non-convex problem, we propose an alternative optimization (AO) algorithm, where penalty and successive convex approximation (SCA) methods are employed. Simulation results validate the effectiveness of the proposed scheme, demonstrating significant performance gains compared to conventional NOMA and OMA benchmarks.

In this paper, the secure information transmission of an energy harvesting (EH) cooperative network is considered, in which multiple eavesdroppers can overhear the forwarded relay signal. To prevent multiple eavesdroppers from decoding confidential signals, the destination transmits the jamming signal while the source transmits the confidential signal to the relay. Simultaneously, the relay can harvest more energy from source signals and destination jamming by the power splitting (PS) protocol, which is depicted as a non-linear EH process. Considering the imperfect self-interference cancellation (SIC) at the destination, secrecy rate maximization optimization is formulated to optimize the transmission power of source and destination. However, the formulated optimization is non-convex. To solve this problem, an iterative algorithm is proposed based on the difference of convex functions (DC) programming, which can transform non-convex optimization problems into successive approximate convex problems. Simulation results show that the proposed algorithm has a quick convergence rate, and the proposed scheme leads to a higher achievable secrecy rate.