PhD , Tongji University, China📖
Haoran Yang is a PhD student in Computer Science at Tongji University, specializing in advanced graph neural networks and multimodal learning. His research since 2020 spans critical topics in artificial intelligence, including graph mining, few-shot learning, and self-supervised learning. His work explores innovative applications of AI in science (AI4Science) and complex graph analysis, advancing understanding and capabilities within these fields.
Google Scholar Profile
Education Background🎓
Haoran Yang is currently a PhD student in Computer Science at Tongji University, where he has been pursuing his research since 2020. His academic journey focuses on advanced topics in graph neural networks, few-shot learning, and multimodal learning, with a strong interest in AI applications for scientific research (AI4Science). Prior to his PhD studies, Haoran completed his undergraduate education in Computer Science, laying the foundation for his deep exploration of AI and machine learning techniques.
Professional Experience🌱
Haoran Yang has gained valuable professional experience as a reviewer for prominent academic journals and conferences, including IEEE Transactions on Knowledge and Data Engineering (TKDE) and the International Conference on Learning Representations (ICLR). His role as a peer reviewer allows him to contribute to the advancement of knowledge in the fields of graph neural networks, few-shot learning, and multimodal AI, ensuring high-quality research is disseminated within the academic community. Additionally, his research experience, focused on graph mining and AI applications in science (AI4Science), has further shaped his expertise in cutting-edge AI methodologies.
Research Interest🔬
- Graph Neural Networks (GNNs): Focused on developing and optimizing GNN architectures, including deep clustering optimization and graph few-shot learning methods.
- Multimodal Learning: Exploring AI4Science and the integration of multiple data sources for enhanced learning outcomes.
- Graph Mining and Self-Supervised Learning: Aiming to refine the application of self-supervised techniques in graph learning to improve model adaptability and performance in varied data environments.
Author MetricsÂ
As a rising researcher, Haoran Yang has contributed significantly to the field of graph neural networks and multimodal AI, with publications and reviews in prestigious venues. His work is gaining recognition, particularly for its innovative approaches to clustering and spectral GNN analysis.
Publications Top Notes 📄
- DCOM-GNN: A Deep Clustering Optimization Method for Graph Neural Networks
- Authors: Haoran Yang, Jiao Wang, Rui Duan, Chao Yan
- Published in: Knowledge-Based Systems, Volume 279, Article 110961, 2023
- Abstract:
This paper presents a novel method, DCOM-GNN, designed to optimize deep clustering in graph neural networks (GNNs). The approach targets improving the clustering performance in graph-based learning by incorporating a clustering loss into the GNN framework. The authors focus on enhancing the quality of learned representations while minimizing intra-cluster distances and maximizing inter-cluster distances, leading to more efficient and accurate graph clustering. The method demonstrates its effectiveness through various experimental results on benchmark datasets, showing improvements over existing clustering techniques in GNNs.
- Key Contributions:
- Introduction of a deep clustering optimization method specifically for GNNs.
- Formulation of a clustering loss that balances intra-cluster and inter-cluster distances.
- Performance evaluation on multiple graph datasets showing the potential of DCOM-GNN in real-world applications.
- Unifying Homophily and Heterophily for Spectral Graph Neural Networks via Triple Filter Ensembles
- Authors: Haoran Yang, et al.
- Conference: NeurIPS 2024 (Accepted)
- Abstract:
This paper proposes a unified framework to address the challenge of integrating homophily (similarity between neighboring nodes) and heterophily (dissimilarity between neighboring nodes) in spectral graph neural networks (SGNNs). The authors introduce Triple Filter Ensembles, a novel technique that combines multiple filter types to account for both homophilic and heterophilic structures in a graph. The method allows SGNNs to adaptively learn from graphs with varying node relationships, improving model robustness and generalization across diverse graph datasets.
- Key Contributions:
- The development of Triple Filter Ensembles to unify homophily and heterophily in SGNNs.
- Enhanced model performance on graph datasets exhibiting mixed node relationships.
- Novel insights into the challenges and solutions for spectral-based GNNs when applied to graphs with both similar and dissimilar neighboring nodes.
Conclusion
Haoran Yang’s research contributions to graph neural networks, particularly in deep clustering optimization and spectral GNNs, represent significant advancements in the field of AI. His innovative work has the potential to transform both theoretical and applied aspects of graph-based learning and multimodal AI. His future directions, focused on AI applications in science and the integration of self-supervised learning and few-shot learning into graph networks, promise to maintain his trajectory as a leading researcher. With continued academic success and the application of his work in real-world contexts, Haoran Yang is a deserving candidate for the Outstanding Research Achievement Award.