Kunyu Zhang

Kunyu Zhang

Deep Learning, AI4Neuroscience and AI4Health

🔥 News

👨‍🔬 About Me

I am an aspiring researcher passionate about harnessing deep learning to drive innovation in healthcare and medicine. My work focuses on applying deep learning to address complex medical challenges and improve patient outcomes.

Currently, I am exploring a range of AI applications in healthcare, including medical image analysis, clinical decision support systems, and predictive modeling for disease diagnosis and treatment. Beyond medicine, I am also committed to developing flexible, efficient, and generalizable deep learning methodologies applicable across diverse scientific domains.

📚 Publications

* Equal contribution

[MICCAI 25] MvHo-IB: Multi-View High-Order Information Bottleneck for Brain Disorder Diagnosis
[Paper] [Code]
Kunyu Zhang, Qiang Li, Shujian Yu
We propose a novel method to capture higher-order interactions while using information bottleneck to reduce redundancy between multi-view information for brain disorder diagnosis.

[Briefings in Bioinformatics] ProtoMol: Enhancing Molecular Property Prediction via Prototype-Guided Multimodal Learning
[Paper] [Code] Yingxu Wang*, Kunyu Zhang*, Jiaxin Huang, Nan Yin, Siwei Liu, Eran Segal
We propose ProtoMol, a prototype-guided multimodal framework that hierarchically aligns molecular graphs and textual descriptions via layer-wise cross-modal attention and a shared prototype space, consistently outperforming prior methods on molecular property prediction while improving interpretability.

[arXiv] Clinical Expert Uncertainty Guided Generalized Label Smoothing for Medical Noisy Label Learning
[Paper] [Code]
Kunyu Zhang, Lin Gu, Liangchen Liu, Yingke Chen, Bingyang Wang, Jin Yan, Yingying Zhu
We address label noise in medical image datasets caused by clinical expert uncertainty in diagnostic notes. We propose a clinical expert uncertainty-aware benchmark and a label smoothing method that significantly outperforms current approaches.

[Neural Networks] Heterogeneity-aware high-efficiency federated learning with hybrid synchronous-asynchronous splitting strategy
[Paper]
Zijian Li*, Boyuan Li*, Kunyu Zhang, Bingcai Wei, Hongbo Liu, Zihan Chen, Xinqiang Xie, Tony Q.S. Quek
We propose a resource-aware adaptive model customization mechanism that mixes synchronous edge feature extraction with asynchronous global classifier updates, achieving faster convergence, higher accuracy, and lower network traffic.

[TIA] Multi-microgrids Peer to Peer Electricity-carbon joint Trading Method Based on Multi-agent Ethical Enhancement Technology
[Paper]
Fashun Shi, Lin Cheng, Yuguang Song, Pengjie Zhao, Yufei Xi, Kunyu Zhang, Haiwang Zhong
We propose a peer-to-peer (P2P) electricity-carbon joint trading method based on multi-agent ethical enhancement technology that combines safe reinforcement learning with a privacy protection module, achieving low-carbon economic operation for multi-microgrids while adhering to “do-no-harm” and “privacy protection” ethical principles.

🔬 Research Interests

  • AI in Biomedicine: Applying artificial intelligence across biomedical domains, including but not limited to neuroscience, medical imaging, clinical diagnosis, and prognosis

  • Machine Learning for Health: Investigating the adaptability and limitations of machine learning methods in biomedical research

  • Generalizable AI Methodologies: Exploring the theoretical foundations for extending AI from biomedical applications to other scientific disciplines through flexible, low-cost, and efficient deep learning frameworks