课程文献中心 更多>>

关键词： Sparse matrices   Hardware acceleration on edge   Gain cell   GC-eDRAM   CAM   Multi-port   Sparse matrix multiplication  

摘要： Sparse General matrix multiplication (SpGEMM) is a fundamental kernel in many scientific and engineering fields, including Artificial Intelligence (AI). However, its intrinsic computation complexity presents substantial challenges, making efficient hardware implementation particularly difficult. This paper proposes SPARCAM, a novel SpGEMM accelerator, developed and optimized for very energy-efficient AI edge applications. SPARCAM is designed using low-power dense Gain Cell embedded DRAM (GC-eDRAM) technology, a processing near memory paradigm, and a modified outer product matrix multiplication algorithm. Despite its quite limited peak theoretical performance, SPARCAM achieves very high energy efficiency due to its low-power architecture and almost 100% utilization of its computing resources. Designed in a commercial 28 nm FDSOI technology, SPARCAM achieves $13.9\times$ speedup over a high-performance embedded CPU when processing large-scale sparse matrices. When multiplying limited-size sparse matrices, SPARCAM obtains $193\times$ speedup over high-performance GPU. SPARCAM reaches about 4.3 orders-of-magnitude, on average, higher energy benefits, and $1892\times$, $181\times$, $2\times$, and $3471\times$, higher energy efficiency (over CPU) compared with state-of-the-art SpGEMM accelerators SpArch, OuterSPACE, MatRaptor, and high-performance GPU, respectively.

关键词： CAD/CAM   Ceramic material   Fretting wear   Interproximal contact loss   Interproximal enamel   Sliding wear  

摘要： This study investigated the fretting and sliding wear behavior of interproximal enamel (IE) against four CAD/CAM ceramic materials including polymer-infiltrated ceramic network (PICN), lithium disilicate (LS), zirconia-reinforced lithium silicate (ZLS), and monolithic zirconia (MZ) were evaluated.

关键词： Closed-form solution   Convex cam profile   Load simulator   Parameters optimization   Wrapping cams  

摘要： Reproducing the aerodynamic torque during the deployment process is essential for the ground verification of the folding wing actuation system. However, the existing load simulators rely on high-power, high-performance hardware and complex control algorithms to achieve rapid tracking of high-magnitude torque, which makes it difficult to quickly deploy in harsh test environments and adapt to different test objects. We present a passive-compliance wrapping-cam load simulator with simple structure, rapid response, and high load capacity and establish an efficient multi-scenario design methodology for diverse load curves. A convexity-embedded cam-profile parametric model is innovatively established during optimization-based design, which preserves satisfactory profile description ability and markedly improved optimization efficiency. A new closed-form solution for wrapping cam synthesis is developed, along with an efficient and versatile hybrid design workflow for the load simulator. The proposed design method was tested using simulations of several representative load cases and was verified on the prototype test bench. Under the sinusoidal target load, the corresponding load simulator achieved good loading accuracy with a relative error of less than 5% and R²>0.99 at various deployment speeds.

关键词： Algorithm-hardware co-design   Deep learning   Intra-row weeding   Lactuca sativa   Mechanical weed control   Precision agriculture  

摘要： Intra-row weed competition severely threatens lettuce (Lactuca sativa L.) cultivation by aggressively consuming soil moisture, nutrients, and light, thereby compromising biomass accumulation and marketable yield. While autonomous mechanical weeding offers a sustainable, chemical-free alternative, its widespread adoption is frequently constrained by prohibitive computational costs and hardware expenses associated with complex robotic systems. To bridge this gap, this study presents a cost-effective intelligent weeding method based on an algorithm-hardware co-design paradigm. On the mechanical front, a novel cam-push rod actuator is designed to replace expensive servo-driven manipulators. Based on a kinematic mathematical model, the cam profile was rigorously optimized to generate a deterministic lateral avoidance trajectory. This "physically encoded" intelligence ensures rapid, precise blade-retraction trajectory tailored to the specific rosette morphology of lettuce seedlings. Synergistically, the system integrates a bespokeightweightdetection model, You Only Look Once-GSBA-Tiny (YOLO-GSBA-Tiny). Through systematic structural pruning and the integration of attention mechanisms, the model reduces parameters and Giga Floating-Point Operations Per Second (GFLOPs) by 69.0 % and 55.6 %, respectively, compared to YOLO 8n, while maintaining a mean average precision of 97.8 %. Controlled environment trials validated the prototype's efficacy, achieving a weed removal rate of 93.86 % with a minimal crop damage rate of 2.08 %. Agronomically, this precision is critical for eliminating competition during the crop’s critical weed-free period, maximizing resource use efficiency, and safeguarding yield potential. Ultimately, this work demonstrates that empowering simple, mathematically optimized mechanical systems with efficient artificial intelligence (AI) provides an agronomically sound and economically scalable pathway for precision weed management.

关键词： 3D Gaussian Splatting   Raster multi-view CAD sketch   Self-supervised learning   Sparse-view reconstruction  

摘要： The automatic reconstruction of 3D computer-aided design (CAD) models from CAD sketches has recently gained significant attention in the computer vision community. Most existing methods, however, rely on vector CAD sketches and 3D ground truth for supervision, which are often difficult to be obtained in industrial applications and are sensitive to noisy inputs. We propose viewing CAD reconstruction as a specific instance of sparse-view 3D reconstruction to overcome these limitations. While this reformulation offers a promising perspective, existing 3D reconstruction methods typically require natural images and corresponding camera poses as inputs, which introduces two major significant challenges: (1) modality discrepancy between CAD sketches and natural images, and (2) difficulty of accurate camera pose estimation for CAD sketches. To solve these issues, we first transform the CAD sketches into representations resembling natural images and extract corresponding masks. Next, we manually calculate the camera poses for the orthographic views to ensure accurate alignment within the 3D coordinate system. Finally, we employ a customized sparse-view 3D reconstruction method to achieve high-quality reconstructions from aligned orthographic views. By leveraging raster CAD sketches for self-supervision, our approach eliminates the reliance on vector CAD sketches and 3D ground truth. Experiments on the Sub-Fusion360 dataset and several cases on other types of CAD sketches demonstrate that our proposed method significantly outperforms previous approaches in CAD reconstruction performance and exhibits strong robustness to noisy inputs. The source code and all related data will be released upon acceptance of the paper.

关键词： Biocompatibility   Cytotoxicity   Dental composite   Thermocycling  

摘要：

• •
  Biological traits of light-cured, CAD/CAM milled, and 3D printed dental composites differ in monomer release and biocompatibility.
• •
  Thermocycling aging reduced cytocompatibility and increased inflammation and oxidative stress in tested dental materials.
• •
  Light-cured composite showed the highest variation in biological properties, especially after extended aging.
• •
  3D printed composite maintained the most stable and favorable biological profile throughout all aging periods.

关键词： Boundary enhancement   CAM   Domain knowledge   Weakly-supervised segmentation  

摘要： For the precise segmentation of surface defects on hot-rolled strip steel, the prevalent issue of weak supervision constitutes a significant bottleneck that limits the advancement of intelligent industrial defect detection *** paper proposes a weakly-supervised segmentation method for surface defects of hot-rolled steel strips (BEDUNet), designed to address the challenges of limited labeled data and ambiguous defect boundaries in industrial defect detection. Due to the difficulties in acquiring high-quality defect data for hot-rolled steel strips and the low contrast between defects and the background, this paper introduces a Boundary Enhancement Module based on domain knowledge (BEDK) and a Dynamic Update Module (DUM) to improve segmentation accuracy. The BEDK module leverages domain knowledge regarding the taxonomic features of hot-rolled steel strip defects (e.g., linearity, blockiness, and boundary vagueness). It employs customized multi-branch convolutional kernels and boundary contrastive learning to enhance the distinction between the defect and background boundaries, thereby generating higher-quality Class Activation Maps (CAMs).The DUM module dynamically adjusts the pseudo-label update cycle and threshold based on model convergence, further elevating the quality of pseudo-labels and overall network *** conducting simulation experiments using a real-world dataset of steel strip defects, the proposed method demonstrated high segmentation accuracy, confirming its effectiveness in simulating industrial production *** reported mean Intersection over Union (mIoU) metrics were 82.7% and 85.1% for the two distinct datasets, as indicated by the experimental evaluation. Furthermore, in practical industrial validation, the accuracy and reliability of the proposed method were further demonstrated, confirming its value in industrial defect segmentation.

关键词： Malaria vaccine   Vaccine hesitancy   Illness perception   Caregivers   Health communication   Ghana   aPR   adjusted Prevalence Ratio   ACTs   artemisinin-based combination therapies   BIC   Bayesian Information Criterion   B-IPQ   Brief Illness Perception Questionnaire   CAM   Complementary and Alternative Medicine   DHIMS2   District Health Information Management System 2   GHS   Ghana Health Service   IQR   Interquartile range   ITNs   Insecticide-Treated Nets   IRS   Indoor Residual Spraying   KMO   Kaiser-Meyer-Olkin   NHIS   National Health Insurance Scheme   SMC   Seasonal Malaria Chemoprevention   VIF   Variance Inflation Factor   WHO   World Health Organization  

摘要： There is a prevalent perception of low personal risk of malaria in Ghana. This perception and other beliefs about the disease held by caregivers may influence hesitancy toward childhood malaria vaccination. Understanding these dynamics is crucial to the national scale-up of the malaria vaccines. This study investigates this relationship to inform strategies for enhancing vaccination coverage.

关键词： Acute myocardial infarction(AMI)   Coronary artery disease(CAD)   Machine learning algorithms   Mitochondrial energy metabolism   Molecular docking   Therapeutic agents  

摘要： Coronary artery disease (CAD) is a leading cause of acute myocardial infarction (AMI), a condition characterized by the sudden blockage of blood flow to the heart muscle, resulting in significant morbidity and mortality worldwide. Despite advances in treatment, early prediction and diagnosis remain challenging. This study aims to construct a predictive model for CAD in AMI patients by focusing on mitochondrial energy metabolism(MME), a critical factor in cardiac function and disease progression. We employed an integrative WGCNA and machine learning algorithm combinations (RF, Lasso and SVM), utilizing 5 distinct CAD and AMI peripheral blood bulk profiles for the elucidation of potential diagnostic biomarkers for CAD patients with AMI. Finally, we discovered that CLPX can be recognized as diagnostic biomarkers, which is related to the molecular and immune pathogenesis of AMI among CAD patients. Meanwhile, we also discovered that MME-related genes can divide CAD and AMI into 2 different molecular subgroups via consensus clustering. Besides, Troglitazone can be regarded as potential therapeutic agents for AMI of CAD patients targeting CLPX via molecular docking. This study underscores the importance of mitochondrial function in cardiovascular diseases and highlights the potential of bioinformatics in advancing precision medicine.

关键词： Partially saturated soils   Cyclic loading   Hydro-mechanical modelling   Hydro-mechanical loading   Historiotropic surface  

摘要： Partially saturated soils are gaining prominence in geotechnical engineering and engineering geology, particularly in the context of extreme climate phenomena. Their intricate response to changing moisture conditions significantly impacts slope stability, foundation design, and infrastructure resilience against climate’s extreme phenomena in geotechnical engineering applications. Current constitutive models mainly focused on predicting monotonic hydro-mechanical (HM) processes, lacking accuracy in predicting soil behaviour under cyclic loading and moisture variations. However, infrastructures like railway tracks impose cyclic loading on the subsoil, while the soil undergoes seasonal moisture variations. This study aims to contribute to this topic by proposing a new HM constitutive model based on the Constitutive Anamnesis Model (CAM) for saturated soils, extending its application to partially saturated conditions. The original CAM effectively predicts the behaviour of fully saturated fine-grained soils under monotonic and cyclic loading, considering soil preloading history. By incorporating partial saturation effects, the extended model offers improved accuracy in simulating the behaviour of low and high plasticity clays under cyclic hydro-mechanical loading, providing a more comprehensive tool for geotechnical engineering applications.