QUICK
MENU
맨위로가기

로고이미지로고이미지

대학장학금입학안내기숙사등록가톨릭대학교대학원입학안내

The Catholic University of Korea

Research Results


Doctoral Student Van Thuy Hoang, from the Department of Artificial Intelligence, Unveils Breakthrough Graph Transformer

  • Writer :관리자
  • Date :2024.02.27
  • Views :360

- Van Thuy Hoang, a doctoral student in the Department of Artificial Intelligence, introduces the cutting-edge Graph Transformer at AAAI 2024.

- Implementation of integrating global and local structural information of graphs based on graph augmentation and self-supervised learning



*Photo Description: Van Thuy Hoang, a doctoral student in the Department of Artificial Intelligence at Catholic University of Korea (supervised by Professor Oh-Joon Lee), presents the Structure-Preserving Graph Transformer model UGT at the international academic conference “AAAI 2024.”





 Van Thuy Hoang, a doctoral student in the Department of Artificial Intelligence at Catholic University (President: Jongchul Won), under the guidance of Professor Oh-Joon Lee, received accolades for a paper on a revolutionary Graph Transformer at the esteemed international academic conference “AAAI 2024”.

 

The Association for the Advancement of Artificial Intelligence (AAAI), established in the United States in 1979, stands as the paramount global authority in the realm of artificial intelligence. Each year, it convenes an international academic conference, inviting leading AI research institutions worldwide and representatives from major corporations to share the latest advancements in AI technology and research achievements. This year marked the 38th AAAI conference, held in Vancouver, Canada, from February 20th for a week.


Van Thuy Hoang, a doctoral student in the Department of Artificial Intelligence at Catholic University of Korea under the mentorship of Professor Oh-Joon Lee, received esteemed recognition for the excellence of his research. He unveiled the Unified Graph Transformer (UGT), a structure-preserving graph transformer model developed by him, at AAAI 2024.


The transformer model is a type of neural network designed to analyze relationships within sequential data, such as words in a sentence, to grasp context and meaning. A graph transformer, which is a modified version of a traditional transformer, allows for the utilization of structured data represented as graphs as input. Essentially, it is a neural network capable of learning not only individual data characteristics but also the relationships between data points.


Previous researches on graph transformers tended to focus either on global or local graph structures, leading to limitations in applications such as community detection and molecular structure analysis. Moreover, these models struggled with node classification tasks where individual node features were crucial.


Hoang successfully addressed these challenges by emphasizing node role similarity information, employing graph augmentation techniques, and integrating local node connectivity information based on self-supervised learning of transition probabilities between nodes. His innovative approach showcased the potential to overcome the limitations of traditional AI research.


Experiments with the newly developed UGT model demonstrated SOTA (state-of-the-art) performance in classifying graphs requiring global feature information, clustering nodes requiring local feature information, and even for node classification tasks. Additionally, the UGT model exhibited an ability to interpret graph structures at a 3-WL level, a capability that previous graph transformer and graph neural network models had not achieved.


Hoang, the doctoral student in the Department of Artificial Intelligence at Catholic University of Korea, expressed hopes that the newly developed UGT model could open new horizons in AI research. He credited Catholic University of Korea for providing excellent education and research programs that enabled him to present his paper at a prestigious international AI conference.



Professor Oh-Joon Lee from the Department of Artificial Intelligence at Catholic University of Korea said, "Van Thuy Hoang, the first doctoral student in the Department of Artificial Intelligence at Catholic University, was highly recognized for his research capabilities at an international academic conference in just his third semester of admission." He added, "The Department of Artificial Intelligence at Catholic University will continue to provide unwavering support for fostering world-class talents in the field of artificial intelligence." (End)