WiFi-based human activity recognition (HAR) has emerged as a focal point within the Internet of Things landscape, owing to its non-intrusive sensing capabilities and inherent privacy-preserving advantages. In existing WiFi-based HAR research, channel state information (CSI) is primarily utilized to capture activity-related features and enable recognition. However, CSI-based cross-domain HAR remains challenged by issues such as redundant subcarriers, limited samples in the target domain, and high sensitivity of CSI to environmental variations. To address these challenges, this paper proposes Wi-DMAR, a WiFi-based cross-domain HAR framework that integrates three key modules. First, an adaptive subcarrier selection module computes the correlation between each subcarrier and the principal components, identifies subcarriers with high contribution, preserves essential activity-related features, reduces data dimensionality, and lowers computational overhead. Second, a conditional diffusion–based data augmentation module employs a Transformer-based feature extractor to capture domain-specific representations of target-domain data, and optimizes domain consistency loss and domain-guided diffusion loss to generate pseudo samples that resemble the target-domain distribution, thereby mitigating sample scarcity. Third, an activity recognition module based on sample similarity learning reformulates the traditional label classification problem into a sample comparison task, by quantifying similarity between samples, it performs activity recognition and enhances cross-domain generalization. Experimental results demonstrate that Wi-DMAR achieves superior recognition accuracy compared with state-of-the-art cross-domain HAR methods such as DiffAR and MetaAct. Ablation studies further confirm that each core component contributes positively to performance improvements.

Sparse Matrix-Matrix Multiplication (SpMM) plays a critical role in high-performance computing and applications like Graph Neural Networks (GNNs).However, due to the sparsity and irregularity of real-world data, optimizing SpMM performance on modern GPUs has remained a significant challenge.Existing methods often involve trade-offs between load balancing and hardware utilization, making it difficult to efficiently handle both long and short rows in sparse matrices.To address these issues, we propose HR-SpMM, a lightweight framework based on adaptive row partitioning and hybrid kernel design.HR-SpMM divides sparse matrix rows into two categories: long rows and short rows, leveraging Tensor Cores and CUDA Cores, respectively, to optimize computational efficiency.Long rows are further partitioned into fixed-size blocks to fully align with the hardware characteristics of Tensor Cores, while short rows adopt a flexible

SimpleConvolution is the most important and time-consuming part of convolutional neural networks (CNN) for image processing. Each slide of the window in two-dimensional convolution will cause repeated memory access. To solve above problems, we propose an algorithm integrated with the hardware level, parallel in kernel Convolution(PIK-Convolution), which reduces the granularity of a SingleConvolution and avoid the problem of storage access pressure caused by spatial locality. Unlike Image to column(im2col) algorithm, we avoid the problem of non-adjacent data in memory directly from hardware perspective. With MGPUSim, a multi-GPU architecture platform, we implement our algorithm. A large number of experimental results show that the proposed algorithm performs well on the new architecture. Compared with the unoptimized algorithm, the overall efficiency of the new algorithm is optimized by 2.2×-7.45×, and the memory access efficiency is an average of 30× of the unoptimized algorithm. Compared to the state-of-the-art convolution acceleration, our algorithm has a performance optimization of 1.5×-2.2.

How to coordinate the design of sampling and Sparse-dense Matrix Multiplication (SpMM) is important in Graph Neural Network (GNN) acceleration. However, existing methods have an imbalance between accuracy and speed in performing GNN inference tasks due to irrational sampling strategies. To solve this problem, we propose an adaptive edge sampling strategy SpMM kernel. It considers the relationship between the number of non-zero elements in each matrix row and the shared memory width. The edge sampling scheme is adaptively selected according to the different situations of each row. Our method reduces the graph size by adaptive edge sampling to fit into the GPU’s shared memory, which decreases the computational cost and increases the data locality ultimately achieving a balance between accuracy and speed in GNN inference. We conducted experiments on NVIDIA RTX 4060 Ti GPU using representative GNN model and datasets. Experimental results show that our designed kernel outperforms the cuSPARSE SpMM kernel and GE-SpMM by up to 20.2× and 17.3× respectively, with less than 1% accuracy loss. Compared to ES-SpMM, it reduces the accuracy loss by 4.7% on average and achieves an average 1.26× speedup.

In future 6G vehicular networks, precise positioning is essential for improving communication quality and efficiency. This paper proposes a novel TDOAIFDOA-based cooperative localization method among multiple vehicles. By selecting anchor vehicles within the range of the base station and utilizing their received echo information, this method enables efficient and low-latency positioning. First, based on a defined variable selection criterion, a subset of vehicles with optimal locations is chosen. Using the echo signals generated by inter-vehicle communication, the method jointly predicts the motion parameters of target vehicles. A time-delay Doppler approach based on matched filtering is employed to estimate the reflected echo information for dynamic vehicle cooperation, ultimately assisting the base station in achieving directional communication with the target vehicles. Results show that under specified noise conditions, the proposed V2V cooperative localization achieves performance close to the CRLB lower bound, with deviations between 0 and 0.8. This method offers a new approach to directional communication in vehicular networks, particularly suited for high-dynamic, high-concurrency large-scale vehicle communication in complex traffic environments.

At Presently, it is still a great challenge to achieve online classification of traffic flows due to the highly varying network environments, e.g., unpredictable new traffic classes, network noise, and congestion. Traditional classification methods work well in stable network environments, but may not exhibit their performance in dynamic environments. To address online classification issues, a granular computing-based classification model (GCCM) is developed, where the spatial and temporal flow granules are defined to make GCCM robust against variations and less sensitive to noise, and the correlations among flow granules are explored to establish the granular relation matrix (GRM). The inherent burst features between packets indicated by GRM prompt GCCM to achieve fine classification in unstable network environments. GCCM analyzes the burst features of packets without inspecting the payload information, and thus can be used to classify encrypted traffic as well as unencrypted traffic at a fast speed. In addition, the GCCM model, depending on difference measurement$D(\cdot)$, is a threshold-based classification, and therefore can be used to distinguish between time-varying classes. The validity of GCCM for online traffic classification is examined through theoretical results. The experimental evaluation of classification for fine and varied classes under dynamic network environments with noise and congestion also demonstrates its superiority in terms of classification accuracy and real-time performance with the state-of-the-art.

Traffic flows with different requirements of quality of service (QoS requirements) are aggregated into different QoS classes to provide differentiated services (Diffserv) and better quality of experience (QoE) for users. The existing aggregation approaches/QoS mapping methods are based on quantitative QoS requirements and static QoS classes. However, they are typically qualitative and time-varying at the edge of the beyond fifth generation (B5G) networks. Therefore, the artificial intelligence technology of preference logic is applied in this paper to achieve an intelligent method for edge computing, called the preference logic based aggregation model (PLM), which effectively groups flows with qualitative requirements into dynamic classes. First, PLM uses preferences to describe QoS requirements of flows, and thus can deal with both quantitative and qualitative cases. Next, the potential conflicts in these preferences are eliminated. According to the preferences, traffic flows are finally mapped into dynamic QoS classes by logic reasoning. The experimental results show that PLM presents better performance in terms of QoE satisfaction compared with the existing aggregation methods. Utilizing preference logic to group flows, PLM implements a novel way of edge intelligence to deal with dynamic classes and improves the Diffserv for massive B5G traffic with quantitative and qualitative requirements.

Internet video traffic exhibits considerable variation as new video services continue to emerge. Some videos require strict real-time performance, while others may aim for a minimal packet loss rate or sufficient bandwidth. Therefore, it is important to develop fine-grained classification mechanisms to realize effective resource management and quality of service (QoS) provisioning. However, the existing methods for classifying video traffic always suffer from two problems: payload inspection and feature selection. In this paper, we propose a novel method that uses fractal characteristics to achieve traffic classification at a fine-grained level. This method requires neither payload signatures nor statistical features. Through rigorous analysis, we prove the feasibility of employing fractal characteristics for video traffic classification and further develop a theoretical framework for the proposed scheme. For the specific scenario of video flow classification, we improve the theory of fractals in terms of estimated spectrum, core domain, segmentation, and threshold setting. The results of an extensive experimental study on several real-world video traffic datasets show that the classification accuracy of the proposed scheme is higher than that of existing methods.

The existing composition approaches focus on either function-oriented composition or QoS-oriented composition. To our best knowledge, there is not a complete solution. Furthermore, existing solutions for QoS-oriented composition are basically a quantitative method. In many domains it is desirable to assess such QoS in a qualitative rather than quantitative way. So we propose a new algorithm, which can implement automatic composition, considering both function and QoS. Moreover this is a qualitative solution. The algorithm is based on reinforcement learning and preference logic reasoning. The theoretical proof and the experiments demonstrate the feasibility and effectiveness of the approach.