- Anomaly Detection Techniques and Applications
- Data Management and Algorithms
- Parallel Computing and Optimization Techniques
- Advanced Neural Network Applications
- Handwritten Text Recognition Techniques
- Sentiment Analysis and Opinion Mining
- Structural Health Monitoring Techniques
- Human Pose and Action Recognition
- Text and Document Classification Technologies
- Infrastructure Maintenance and Monitoring
- Advanced Image and Video Retrieval Techniques
- Advanced Text Analysis Techniques
- Advanced Database Systems and Queries
- ECG Monitoring and Analysis
- Nanomaterials for catalytic reactions
- Metal and Thin Film Mechanics
- Structural Engineering and Vibration Analysis
- 3D Shape Modeling and Analysis
- Video Analysis and Summarization
- Advanced Wireless Communication Techniques
- Acoustic Wave Resonator Technologies
- Recommender Systems and Techniques
- Health Literacy and Information Accessibility
- Concrete Corrosion and Durability
- Smart Materials for Construction
Shandong Jianzhu University
2025
Ping An (China)
2023
Southeast University
2023
National Taipei University of Technology
2022
National Taiwan University
2022
Shaoxing University
2022
Hong Kong University of Science and Technology
2007
University of Hong Kong
2007
Aspect Sentiment Triplet Extraction (ASTE) has become an emerging task in sentiment analysis research, aiming to extract triplets of the aspect term, its corresponding opinion and associated polarity from a given sentence. Recently, many neural networks based models with different tagging schemes have been proposed, but almost all them their limitations: heavily relying on 1) prior assumption that each word is only single role (e.g., or etc. ) 2) word-level interactions treating...
We present GPUQP, a relational query engine that employs both CPUs and GPUs (Graphics Processing Units) for in-memory co-processing. are commodity processors traditionally designed graphics applications. Recent research has shown they can accelerate some database operations orders of magnitude over CPUs. So far, there been little work on how be programmed heavy-duty constructs, such as tree indexes joins, well full-fledged GPU co-processor performs in comparison with their CPU counterparts....
In winter, the ice and snow on asphalt pavement reduce friction coefficient of pavement, which may lead to serious traffic accidents large-scale congestion. Taking preventive measures ensure safety by accurately predicting road surface temperature is an economical environmentally friendly solution. However, (RST) prediction a challenging task due complicated uncertainty periodicity. To improve accuracy RST prediction, this paper aims propose advanced ensemble deep learning model using gated...
Recently, graphics processing units, or GPUs, have become a viable alternative as commodity, parallel hardware for general-purpose computing, due to their massive data-parallelism, high memory bandwidth, and improved programming interface. In this paper, we explore the use of GPU on grid file, traditional multidimensional access method. Considering characteristics design massively multi-threaded GPU-based file static, memory-resident point data. Moreover, propose hierarchical variant handle...
For cable-stayed bridges, cables are very important components to maintain the safety of whole bridge structure. It is well-known that change in cable force reflects health bridge. Therefore, it necessary detect and quantify local damage prior occurrence a failure. To this end, an improved residual algorithm independent static load vector was proposed study. The method mainly makes use particularity only few coefficients vectors nonzero. By combining two different loading modes, new...
Modeling semantic information is helpful for scene text recognition. In this work, we propose to model and visual jointly with a Visual-Semantic Transformer (VST). The VST first explicitly extracts primary from feature maps transformer module visual-semantic alignment module. then joined the (viewed as sequence) form pseudo multi-domain sequence combining information, which subsequently fed into an transformer-based interaction enable learning of interactions between features. way, features...
Efficient matrix operations have been deemed keys to efficient data analysis. Dual-Triangular QR Decomposition (DT-QRD) is a critical component in Tall and skinny decomposition (TS-QRD), which widely-used operation with various applications, such as compression feature extraction. In order accelerate DT-QRD, this paper, we propose new acceleration framework, including Global Acceleration Schemes, Partially <tex xmlns:mml="http://www.w3.org/1998/Math/MathML"...
In the mechanical analysis of steel structures, whether it is static or dynamic analysis, necessary to establish structural stiffness matrix first. process building matrix, same element usually has different node code connection orders, and never been argued orders will have an effect on matrix. this study, influence difference in order construction studied. First, global coordinate system established when different. It found that indeed inconsistent for with orders. elements are extracted...
Aspect Sentiment Triplet Extraction (ASTE) has become an emerging task in sentiment analysis research, aiming to extract triplets of the aspect term, its corresponding opinion and associated polarity from a given sentence. Recently, many neural networks based models with different tagging schemes have been proposed, but almost all them their limitations: heavily relying on 1) prior assumption that each word is only single role (e.g., or etc. ) 2) word-level interactions treating...
Benchmarking of 3D Shape retrieval allows developers and researchers to compare the strengths different algorithms on a standard dataset. Here we describe procedures involved in developing benchmark issues involved. We then discuss some current shape benchmarks efforts our group others. also review performance evaluation measures that are developed used by community. After give an overview contest (SHREC) tracks run under EuroGraphics Workshop Object Retrieval details organized for SHREC...
Filter pruning is widely adopted to compress and accelerate the Convolutional Neural Networks (CNNs), but most previous works ignore relationship between filters channels in different layers. Processing each layer independently fails utilize collaborative across In this paper, we intuitively propose a novel method by explicitly leveraging Filters Similarity Consecutive Layers (FSCL). FSCL compresses models whose corresponding features are more worthless model. The extensive experiments...
Filter pruning is widely adopted to compress and accelerate the Convolutional Neural Networks (CNNs), but most previous works ignore relationship between filters channels in different layers. Processing each layer independently fails utilize collaborative across In this paper, we intuitively propose a novel method by explicitly leveraging Filters Similarity Consecutive Layers (FSCL). FSCL compresses models whose corresponding features are more worthless model. The extensive experiments...
High quality semipolar (1122) AlN films have been grown on (1010) m -plane sapphire substrates with the help of dual moderate-temperature-grown (MTG) interlayers by using metal-organic chemical vapor deposition technology. The layer thickness film was determined employing relative optical transmittance spectrum measured ultraviolet-visible spectrophotometer. effect insertion 80 nm-thick MTG interlayer structural and properties investigated in detail based characterization results atomic...
The concept of difference and sum co-array(DSCA) has become a new design idea for planar sparse arrays. Inspired by the shifting invariance property DSCA, specific configuration named here as improved L-shaped array is proposed. Compared to other traditional 2D configurations such nested arrays hourglass arrays, proposed configuration larger central consecutive ranges in its thus increasing DOF. At same time, mutual coupling effect also reduced due enlarged spacing between adjacent sensors....
Conventional event detection models under supervised learning settings suffer from the inability of transfer to newly-emerged types owing lack sufficient annotations. A commonly-adapted solution is follow a identify-then-classify manner, which first identifies triggers and then converts classification task via few-shot paradigm. However, these methods still fall far short expectations due to: (i) insufficient discriminative representations in low-resource scenarios, (ii) trigger...
Contrastive learning has emerged as a powerful tool for graph representation learning. However, most contrastive methods learn features of graphs with fixed coarse-grained scale, which might underestimate either local or global information. To capture more hierarchical and richer representation, we propose novel Hierarchical Learning (HCL) framework that explicitly learns in manner. Specifically, HCL includes two key components: adaptive to Pool (L2Pool) method construct reasonable...