Welcome to Gareth Qiming’s Homepage
Yupeng HeTel: +86 8208208820 Email: he_yupeng@tju.edu.cn |
|
Education
School of Mathematics, Tianjin University, 2018.09-present
Bachelor of Science - Mathematics and Applied Mathematics
RESEARCH EXPERIENCE
Reinforcing Graph Representation via Multi-view Pooling for Relation Extraction
Second Author, Research paper (ACL 2022 Conference Under Review), 2021/08-2021/10
Achievement: A model DA-GDPN (Dual-channel Attention Graph Dynamic simplified Pooling Network) is proposed to capture the relation between long-distance words in balance and reduce noise. Specifically, a multilayer topology learning method based on multi-view sifting is proposed to iteratively learn the underlying graph structure among tokens. Besides, a dual-channel self-attention is also presented to capture the relevance between tokens.
Contribution: Initialized the edges of latent multi-view graph and discovered complete indicative latent graphs through Learnable Graph Generator (LGG); proposed Dynamic Simplified Pooling (DSPool), which is a simplified multi-view graph pool technology, to refine latent multi-view graphs; presented a Dual-channel self-Attention (DAttention) to reconstruct latent multi-view graph.
Modeling Variable Space with Residual Tensor Network for Multivariate Time Series
Third Author, Research paper (submitted), 2021/04-2021/09
Achievement: Based on uniform MPS network and Series-variables self-attention mechanism, the project tends to model multivariate time series which will be potentially applied to long series tasks.
Contribution: Combined the U-MPS model and residual network, replaced the traditional attention mechanism by Series-variables self-attention mechanism in this multivariate time series prediction task, and compared the proposed model with existing baselines by metrics.
Visibility Graph Regression for Pressure Peak Prediction in Fracturing Processes
Second Author, Research paper (submitted), 2021/01-2021/08
Achievement: The paper describes the methodology of constructing a network with attribute graphs for sequence data, and presents reasonable prediction schemes and automatic optimization methods for graph networks.
Contribution: Compared graph neural network prediction methods with traditional machine and deep learning regression prediction, evaluated all comparison methods by metrics, and visualized the prediction results in order to represent the trend of the prediction.
AWARDS
Second Prize (Tianjin), Contemporary Undergraduate Mathematical Contest, 2020.10
Successful Participant, Mathematical Contest in Modeling, 2021.02
ACTIVITIES
Vaccination Volunteer
Volunteer service team of Tianjin university, 2021.04
Fruit Project Distribution Volunteer
Union of student financial assistance of volunteer service in China-Tianjin university, 2021.05
Volunteer Activity for Leading New Students
Volunteer service team, School of mathematics, Tianjin University, 2021.08
TOEFL AND GRE
TOEFL: 89
GRE: 322