Research Interests

Interpretable Graph Learning

Interpretable graph learning on heterogeneous graph can be applied in different graph-relavent tasks, such as node classification, graph classification, and link prediction. It involves applying machine learning techniques to complex graphs containing diverse node and edge types, with a focus on explaining model predictions. This entails feature extraction from heterogeneous data, the use of interpretable models like decision trees or rule-based systems, and the incorporation of explainability methods such as feature importance scores and attention mechanisms. Understanding cross-type relationships and ensuring scalability are key challenges, while applications span recommendation systems, fraud detection, biology, and knowledge graph reasoning. Ethical considerations, such as bias detection and fairness, are integral to responsible heterogeneous graph interpretation, making it essential for transparent and accountable decision-making in diverse real-world contexts.

Multimodal Mining

Multimodal mining encompasses the analysis and machine learning applied to heterogeneous data sources, which can include text, images, audio, and more. Its primary objective is to extract valuable insights by seamlessly integrating and analyzing these diverse modalities, employing techniques such as feature fusion and the investigation of intermodal relationships. Its versatile applications extend to computer vision, natural language processing, healthcare, and autonomous systems, though it presents challenges such as data alignment, dimensionality reduction, and ethical considerations. Particularly, I focus on image captioning task which aims to generate the natural language for the image. Differenet from existing tasks, we try to generate caption for context-aware multiple images.