Given the characteristics of road defect detection, such as the small proportion of defective targets in defective images, the large scale difference of different types of defects, and the complex background environment, as well as the limitations of current methods, which are unsuitable for edge deployment and real-time detection, we propose a road defect detection model with focused feature enhancement and multi-scale feature fusion. Firstly, we create a Global-Local Feature Extractor (GLFE) that extracts global and local feature information from road defect images, allowing us to adapt to the vast span and slender features of crack objects while also improving feature extraction capacity for low resolution images. Secondly, we propose an Adaptive-Generalized Feature Pyramid Network (Adaptive-GFPN) based on contextual information as the model's feature fusion network, as well as design and introduce a Context Aware Fusion Module (CAFM) and a dynamic up-sampling operation (DYsample) that can be guided and adapted based on the contextual information of features at different scales to improve defect detection accuracy in complex backgrounds. Finally, to improve detection efficiency, we design a Lightweight Detection Head (LDH) that reduces the number of network parameters and computations. The experimental results reveal that, when compared to the baseline model YOLOv8, our model improved [email protected] by 5.2%, Precision by 3.5%, and Recall by 4.5% on the RDD2022 dataset.

Due to the escalating demand to analyze large graphs, many organizations are now collecting billion-level property graph datasets, concurrently executing many complex graph queries against them, and expecting interactive-level response latency. However, such requirements are particularly challenging because of the notoriously irregular data access pattern and complex dependencies between heterogeneous subtasks. Despite the widespread availability of many-core CPUs and high-speed networking in modern datacenters, existing distributed graph query systems struggle with their inherent inefficiencies, resulting in low hardware utilization and poor query performance on these state-of-the-art hardware. To address these challenges, we introduce the Partitioned Stateful Traversal Machine (PSTM), which extends the Gremlin graph traversal machine. PSTM retains the expressive power of the Gremlin query language, enabling it to accommodate a wide range of graph query tasks, including traversal, pattern matching, filtering, and result aggregation. It additionally introduces query memoranda, allowing for more efficient implementation and execution of numerous graph queries in distributed environments. Moreover, PSTM facilitates various system-level optimizations, such as massively parallel execution, overlapping computation with communication, locality-aware data access, and lightweight progress tracking. Building upon PSTM, we develop GraphDance, a distributed graph database featuring an efficient asynchronous PSTM run-time. Our evaluations, conducted on an 8-node cluster, show that GraphDance achieves millisecond-level query latency for complex queries on terabyte-scale graphs, with an average latency reduction of 89.2% across all interactive complex queries in the LDBC SNB benchmark compared to existing distributed graph query systems.

Due to the sparse nature of Mixture-of-Experts (MoE) models, they are particularly suitable for hybrid CPU/GPU inference, especially in low-concurrency scenarios. This hybrid approach leverages both the large, cost-effective memory capacity of CPU/DRAM and the high bandwidth of GPU/VRAM. However, existing hybrid solutions remain bottlenecked by CPU computation limits and CPU-GPU synchronization overheads, severely restricting their ability to efficiently run state-of-the-art large MoE models, such as the 671B DeepSeek-V3/R1.

Addressing the problem of complex communication in intelligent transportation system (ITS) under aggregated interference, it is clear that reconfigurable intelligent surface (RIS)-assisted communication is an advantageous solution to the problem of high energy consumption, due to its low-cost and sustainable features. We first propose a transmission protocol architecture of dedicated short range communication (DSRC) compatible mode under cellular-vehicle to everything (C-V2X), and a transmission selection scheme, which selects whether to perform RIS-assisted communication or not based on the requirement of real-time outage probability. Next, single-RIS-assisted and multi-RIS-assisted ITS models are proposed separately, and precise expressions for outage probability, bit error rate, and channel capacity are successfully derived. Finally, we provide sufficient Monte Carlo simulations and analysis of the proposed systems, and the simulation results verified the effectiveness of the proposed algorithm and the correctness of the theoretical analysis.

In this paper, we propose two vehicular communication system models based on Reconfigurable Intelligent Surface (RIS)-assisted sources. Specifically, we assume two scenarios: RIS-Relay aided RIS-assisted source and Relay aided RIS-assisted source, where the relay employs the Power Splitter (PS) to divide the received signal into information transmission and energy harvesting parts. Moreover, both scenarios consider whether a direct link exists. We derive in detail precise approximate closed-form expressions for the outage probability in various vehicular communication scenarios under Nakagami-m channels, and conduct asymptotic analysis of the outage probability to gain further insights. Finally, we validate all the analytical results through simulations, demonstrating the accuracy of the expressions we derived. The simulation results indicate that system and channel parameters significantly influence system performance. Additionally, through comparison, we find that the system with RIS-assisted source and RIS-Relay outperforms system where the RIS is used solely as transmitter or relay, as well as the system with RIS-assisted source and energy harvesting relay.

Click-through rate prediction plays a crucial role in contemporary recommender systems, which aim to refine recommended items by predicting the click-through rate to enhance the overall recommendation effectiveness. However, existing click-through rate prediction models still do not fully leverage both long-term and short-term user behavior information. Consequently, this paper proposes a novel click-through rate prediction model, named Interest Evolutionary Network with Hash and Multiple Attention Network (IEHAN). The model categorizes user behaviors into long-term and short-term categories, employing a hash sampling method to model long-term user behaviors. By directly filtering behavioral items with the same hash signature as the target item from the entire sequence, the IEHAN model effectively models long-term user interests, thereby reducing information loss. Additionally, the model employs the multi-head attention mechanism and the GRU model with an attention update gate to integrate the relevance of the user's interest state and the target item, thereby enhancing the impact of relevant interests on interest evolution and shaping short-term user interest. Experimental results demonstrate that the IEHAN model achieves higher AUC and lower LOSS evaluation metrics on the Taobao dataset and the Books subset of the Amazon dataset, surpassing other existing models and exhibiting superior prediction accuracy.

Many real-time OLAP systems have been proposed to query evolving data with sub-second latency. Although this feature is highly attractive, it is very hard to be achieved on analytic graph queries that can only be answered after accessing every connected vertex. Fortunately, researchers recently observed that answering pairwise queries is enough for many real-world scenarios. These pairwise queries avoid the exhaustive nature and hence may only need to access a small portion of the graph. Obviously, the crux of achieving low latency is to what extent the system can eliminate unnecessary computations. This pruning process, according to our investigation, is usually achieved by estimating certain upper bounds of the query result in existing systems.

Silicon photonics technologies are a particularly attractive solution for developing low-cost optical interconnects with high performance. Imec is developing a silicon photonics technology platform. Developing this platform requires continuous process optimization and design verification, both of which are enabled by the flexible wafer-level test solution that is presented in this paper. The test station enables semi-automatic optical and electro-optical testing of passive and active silicon photonics components and circuits, including waveguides, fiber grating couplers, photodetectors, modulators, filters etc. The measured insertion loss of fiber grating couplers is repeatable to within 0.07dB (6σ), for photodetector responsivity the repeatability is around 0.02A/W (6σ). Calibration procedures have been designed to ensure the long-term reproducibility of measurement results. This is demonstrated with wafer-level measurement data for fiber grating couplers and photodetectors that were gathered over a five-month period. The reproducibility over this period is 0.8dB for the insertion loss and 0.09A/W for the responsivity measurement.

Eye pupil localization is an important part in computer vision applications such as face recognition, gaze estimation and so on. In this paper, we propose an improved method for precise and fast eye pupil localization. Based on gradient boosting decision tree(GBDT) algorithm, a more accurate localization is achieved by increasing the weight of the training samples with larger errors in a moderate rate. Furthermore, a pruning strategy is utilized to avoid overfitting and reduce the localization time without accuracy loss. Experimental results show that the improved method achieves an accuracy of 92.39% at a speed as fast as 1.7ms to locate in the range of eye pupil on BioID database. The proposed method outperforms most state-of-the-art methods in terms of localization accuracy and consumed time.

In order to meet the diverse requirements of market, the architecture of rapid development system for high-speed trainsets body is proposed firstly. A three-layered skeleton model which captures important criteria and knowledge is built so that the chief engineer, system designers and structure engineers can be authorized different rights to carry on development activities respectively. To guarantee the consistency of design change at all levels, the feature control strategy based on multiple templates is employed after intensive analysis of relationships of parameters, information transmitting and inheriting mechanisms of geometry features. The template control process for the body with 8 windows is introduced in detail. Then implementation methods of calling the self-defined control function and nesting of conditional expressions are compared in terms of the efficiency, readability, stability and improvability of the system. The proposed system is verified and the result shows that a new body model can be regenerated very quickly by input parameters through the system interface.

This letter develops a novel method for fingerprint singular point detection based on a new singularity representation of ridge-valley region called orientation entropy. The candidate singular point is obtained by the multiple-scale analysis of orientation entropy and some post processing steps are proposed to filter the spurious core and delta points. An iteration compensation scheme is proposed to search the precise location for core points against the offset further. Performance of the proposed method has been evaluated on the dataset of FVC2002 DB1. Experimental results show that the multiple-scale orientation entropy is correct and effective for singular detection and the location compensation scheme reduces the distance between the detection result and the truth singular point.