A5100405151- Computational Drug Discovery Methods
- Video Surveillance and Tracking Methods
- Advanced Image Processing Techniques
- Sparse and Compressive Sensing Techniques
- Advanced Vision and Imaging
- Image Processing Techniques and Applications
- Image and Signal Denoising Methods
- Gait Recognition and Analysis
- Anomaly Detection Techniques and Applications
- Parallel Computing and Optimization Techniques
- Natural Language Processing Techniques
- Face recognition and analysis
- Human Pose and Action Recognition
- Brain Tumor Detection and Classification
- Topological and Geometric Data Analysis
- Advanced Sensor and Control Systems
- Fault Detection and Control Systems
- Ubiquitin and proteasome pathways
- Tensor decomposition and applications
- Machine Fault Diagnosis Techniques
- Time Series Analysis and Forecasting
- Machine Learning in Materials Science
- Surgical Simulation and Training
- Computer Graphics and Visualization Techniques
- Advanced Graph Neural Networks
University of Nottingham Malaysia Campus
2023-2025
Hunan University
2024
Hubei University of Technology
2024
Dalian University of Technology
2024
Xidian University
2024
The University of Tokyo
2023
Tsinghua University
2022
Xinxiang Medical University
2022
Google (United States)
2020-2021
Beijing Technology and Business University
2021
Abstract Motivation Accurately predicting molecular metabolic stability is of great significance to drug research and development, ensuring safety effectiveness. Existing deep learning methods, especially graph neural networks, can reveal the structure drugs thus efficiently predict molecules. However, most these methods focus on message passing between adjacent atoms in graph, ignoring relationship bonds. This makes it difficult for estimate accurate representations, thereby being limited...
In this work, we present two parallel algorithms for the large-scale discrete Fourier transform (DFT) on Tensor Processing Unit (TPU) clusters.The are associated with DFT formulations: one formulation, denoted as KDFT, is based Kronecker product; other famous Cooley-Tukey algorithm and phase adjustment, FFT.Both KDFT FFT formulations take full advantage of TPU's strength in matrix multiplications.The formulation allows direct use nonuniform inputs without additional step.In algorithms, same...
Lysine-specific demethylase 1 (LSD1) is a histone-modifying enzyme, which significant target for anticancer drug research. In this work, 40 reported tetrahydroquinoline-derivative inhibitors targeting LSD1 were studied to establish the three-dimensional quantitative structure–activity relationship (3D-QSAR). The established models CoMFA (Comparative Molecular Field Analysis (q2 = 0.778, Rpred2 0.709)) and CoMSIA Similarity Index 0.764, 0.713)) yielded good statistical predictive properties....
Abstract Motivation Accurately predicting the degradation capabilities of proteolysis-targeting chimeras (PROTACs) for given target proteins and E3 ligases is important PROTAC design. The distinctive ternary structure PROTACs presents a challenge to traditional drug–target interaction prediction methods, necessitating more innovative approaches. While current state-of-the-art (SOTA) methods using graph neural networks (GNNs) can discern molecular proteins, thus enabling efficient PROTACs’...
In this work, we present two parallel algorithms for the large-scale discrete Fourier transform (DFT) on Tensor Processing Unit (TPU) clusters. The are associated with formulations of DFT: one is based Kronecker product, to be specific, dense matrix multiplications between input data and Vandermonde matrix, denoted as KDFT in work; other famous Cooley-Tukey algorithm phase adjustment, FFT work. Both take full advantage TPU's strength multiplications. formulation allows direct use nonuniform...
Most existing crowd counting methods have focused on pure convolutional neural network based supervised algorithms. Although these attained good results some datasets, they still encounter several common problems. The cost of labeling annotations for is huge and the shortage labeled datasets limits further development algorithms counting. Meanwhile, CNN-based certain limitations in building connections among features. To overcome those problems, we proposed a semi-supervised algorithm that...
Fault diagnosis in industrial production is vital as emerging technologies require innovative methods to identify subtle fault distinctions. Traditional machine learning approaches for steel plate classification inadequately exploit sample relationships, limiting accurate diagnosis. Thus, we proposed use graph construction conjunction with Graph Attention Networks (GAT) classification. We introduced the following four techniques generate adjacency matrices representing connections: k-nearest...
Multiobjective optimization problems (MOPs) are prevalent in machine learning, with applications multi-task learning under fairness or robustness constraints, etc. Instead of reducing multiple objective functions into a scalar objective, MOPs aim to optimize for the so-called Pareto optimality set which involves optimizing more than one function simultaneously, over models thousands / millions parameters. Existing benchmark libraries mainly focus on evolutionary algorithms, most zeroth-order...
Contrastive Language-Image Pre-training (CLIP) models excel in integrating semantic information between images and text through contrastive learning techniques. It has achieved remarkable performance various multimodal tasks. However, the deployment of large CLIP is hindered resource-limited environments, while smaller frequently fail to meet benchmarks required for practical applications. In this paper, we propose a novel approach, ComKD-CLIP: Comprehensive Knowledge Distillation...
Graph coarsening is a technique for solving large-scale graph problems by working on smaller version of the original graph, and possibly interpolating results back to graph. It has long history in scientific computing recently gained popularity machine learning, particularly methods that preserve spectrum. This work studies from different perspective, developing theory preserving distances proposing method achieve this. The geometric approach useful when with collection graphs, such as...
Current Scene text image super-resolution approaches primarily focus on extracting robust features, acquiring information, and complex training strategies to generate images. However, the upsampling module, which is crucial in process of converting low-resolution images high-resolution ones, has received little attention existing works. To address this issue, we propose Pixel Adapter Module (PAM) based graph pixel distortion caused by upsampling. The PAM effectively captures local structural...
Since a great number of devices in everyday life are susceptible to damage and failure, the failure large can be serious threat human safety as well costly property damage, it is important achieve real-time, effective, accurate fault analysis equipment. Traditional machine learning methods, however, ignore spatio-temporal dependencies industrial equipment data, which affects efficiency accuracy detection. In this regard, we introduce hybrid LSTM-WDCNN (Long Short-Term Memory Network - Deep...