A5101665003- Advanced Graph Neural Networks
- Complex Network Analysis Techniques
- Traffic Prediction and Management Techniques
- Transportation Planning and Optimization
- Graph Theory and Algorithms
- 3D Surveying and Cultural Heritage
- Traffic control and management
- Time Series Analysis and Forecasting
- Advanced Neural Network Applications
- Robotics and Sensor-Based Localization
- Recommender Systems and Techniques
- Building Energy and Comfort Optimization
- Power Systems Fault Detection
- Advanced Memory and Neural Computing
- Machine Learning in Materials Science
- Magneto-Optical Properties and Applications
- 2D Materials and Applications
- Advanced Algorithms and Applications
- Human Pose and Action Recognition
- Brain Tumor Detection and Classification
- Artificial Intelligence in Healthcare
- BIM and Construction Integration
- Neural Networks and Applications
- Advanced Image and Video Retrieval Techniques
- Robotic Path Planning Algorithms
Ministry of Agriculture and Rural Affairs
2024-2025
Peking University
2020-2024
Huazhong Agricultural University
2021-2024
Beijing Jiaotong University
2024
Universidad del Noreste
2020
University of Massachusetts Lowell
2011-2012
Spatial-temporal forecasting has attracted tremendous attention in a wide range of applications, and traffic flow prediction is canonical typical example. The complex long-range spatial-temporal correlations bring it to most intractable challenge. Existing works typically utilize shallow graph convolution networks (GNNs) temporal extracting modules model spatial dependencies respectively. However, the representation ability such models limited due to: (1) GNNs are incapable capture...
The "Graph pre-training and fine-tuning" paradigm has significantly improved Graph Neural Networks(GNNs) by capturing general knowledge without manual annotations for downstream tasks. However, due to the immense gap of data tasks between fine-tuning stages, model performance is still limited. Inspired prompt in Natural Language Processing(NLP), many endeavors have been made bridge graph domain. But existing methods simply reformulate form ones. With premise that graphs are compatible with...
Manual annotation of piglet imagery across varied farming environments is labor-intensive. To address this, we propose a semi-automatic approach within an active learning framework that integrates pre-annotation model for detection. We further examine how data sample composition influences efficiency to enhance the deployment lactating detection models. Our study utilizes original samples from pig farms in Jingjiang, Suqian, and Sheyang, along with new Yinguang farm Danyang. Using YOLOv5...
Despite the recent success of Message-passing Graph Neural Networks (MP-GNNs), strong inductive bias homophily limits their ability to generalize heterophilic graphs and leads over-smoothing problem. Most existing works attempt mitigate this issue in spirit emphasizing contribution from similar neighbors reducing those dissimilar ones when performing aggregation, where dissimilarities are utilized passively positive effects ignored, leading suboptimal performances. Inspired by idea attitude...
Traffic flow forecasting is of great significance for improving the efficiency transportation systems and preventing emergencies. Due to highly non-linearity intricate evolutionary patterns short-term long-term traffic flow, existing methods often fail take full advantage spatial-temporal information, especially various temporal with different period shifting characteristics road segments. Besides, globality representing absolute value status indicators locality relative have not been...
While Graph Neural Networks (GNNs) have achieved remarkable results in a variety of applications, recent studies exposed important shortcomings their ability to capture heterogeneous structures and attributes an underlying graph. Furthermore, though many Heterogeneous GNN (HGNN) variants been proposed state-of-the-art results, there are limited theoretical understandings properties. To this end, we introduce graph kernel HGNNs develop Kernel-based (HGK-GNN). Specifically, incorporate the...
Graph representation learning aims to effectively encode high-dimensional sparse graph-structured data into low-dimensional dense vectors, which is a fundamental task that has been widely studied in range of fields, including machine and mining. Classic graph embedding methods follow the basic idea vectors interconnected nodes can still maintain relatively close distance, thereby preserving structural information between graph. However, this sub-optimal due to: (i) traditional have limited...
Graph neural networks (GNNs) have achieved significant success in numerous fields under settings where training and testing graphs are identically distributed. However, this setting is rarely satisfied real life. Due to the lack of out-of-distribution (OOD) generalization abilities, existing GNNs methods perform disappointingly when there exist distribution shifts between graphs. Though several attempts been made deal with issue, they mainly focus on structural properties while overlooking...
With the increasing construction of ice arena facilities, addressing their energy consumption issues has become crucial, emphasizing need for renewable utilization. This study aims to determine contribution rate photovoltaic (PV) power generation in indoor arenas across different climate zones China and proposes corresponding PV application strategies. By modeling vertical zoning above arena, six typical cities was simulated, calculating annual monthly rates proposing utilization The...
Large-scale "pre-train and prompt learning" paradigms have demonstrated remarkable adaptability, enabling broad applications across diverse domains such as question answering, image recognition, multimodal retrieval. This approach fully leverages the potential of large-scale pre-trained models, reducing downstream data requirements computational costs while enhancing model applicability various tasks. Graphs, versatile structures that capture relationships between entities, play pivotal...
The foundation of robot autonomous movement is to quickly grasp the position and surroundings robot, which SLAM technology provides important support for. Due complex dynamic environments, single-sensor methods often have problem degeneracy. In this paper, a multi-sensor fusion method based on LVI-SAM framework was proposed. First all, state-of-the-art feature detection algorithm SuperPoint used extract points from visual-inertial system, enhancing ability in scenarios. addition, improve...
MicroRNA (miRNA) has became an increasingly important class of attractive drug targets in recent studies. However, there are only few computational tools aiming to predict drugmi-RNA resistance associations. Hence, it is great significance develop effective and high accuracy methods for predicting In this work, we propose a novel method abbreviated as "DMR-GCN", which enhances interaction prediction by using layer attention graph convolution network multi channel feature extraction....
Traditional visual place recognition (VPR) methods generally use frame-based cameras, which is easy to fail due dramatic illumination changes or fast motions. In this paper, we propose an end-to-end network for event can achieve good performance in challenging environments. The key idea of the proposed algorithm firstly characterize streams with EST voxel grid, then extract features using a convolution network, and finally aggregate improved VLAD realize streams. To verify effectiveness...
Ever-growing CNN size incurs a significant amount of redundancy in model parameters, which turn, puts considerable burden on hardware. Unstructured pruning is widely used to reduce sparsity. While, the irregularity introduced by unstructured makes it difficult accelerate sparse CNNs systolic array. To address this issue, variety accelerators have been proposed. SIGMA, state-of-the-art GEMM accelerator, achieves speedup over However, SIGMA suffers from two disadvantages: 1) only supports...