Ken Chen
Thesis Title
Contrastive Self-supervised Learning in Graphs
Research overview
Contrastive self-supervised learning learns the representations through maximizing
feature agreement with views generated by a variety of data augmentation strategies.
Pretext tasks and data augmentation are essential in contrastive learning. However,
researches on well-designed pretext tasks and efficient data augmentation strategies for
complex graph data are limited.
This study aims to develop pretext tasks and data augmentation strategies in graph.
Therefore, the experiment is to train proposed methods in semi-supervised and
unsupervised manners, and then to evaluate them and to compare them with
representative baseline methods on different downstream tasks.
Supervisors
Qualifications
B.Eng. Coastal Engineering, Zhejiang University, China (2019)
M.Eng. Energy Engineering, Zhejiang University, China (2022)