Yao Hu

Yao Hu is currently a Research Assistant Professor with the Department of Data Science and Artificial Intelligence (DSAI) at The Hong Kong Polytechnic University. Prior to that, he was a Postdoctoral Fellow of DSAI under the supervision of Prof. Kay Chen Tan and Dr. Jibin Wu, as a member of the Machine Intelligence and Nature-Inspired Computing (MIND) Lab. He received the B.S. degree in mining engineering and the M.Sc. degree in control science and engineering from the China University of Mining and Technology under the supervision of Prof. Xiaoyan Sun, and the Ph.D. degree at the City University of Hong Kong under the supervision of Prof. Kay Chen Tan and Prof. Linqi Song.

His current research focus is on the vision-language model training, data scarcity problem, privacy-preserving domain adaptation, federated learning, and applied deep learning for medical image analysis.

Email  /  Google Scholar  /  Github  /  CV

Research

This work proposes a source-free semi-supervised transfer learning framework for diagnosing mental disorders using fMRI scans, addressing the challenges of limited labeled data and cross-site data privacy in medical imaging.

This work focuses on communication-efficient federated learning, proposing a novel framework to handle heterogeneous structured client models, reducing communication overhead while maintaining model performance in distributed learning scenarios.

This work presents a federated multi-task learning framework for jointly diagnosing multiple mental disorders from MRI scans, enabling collaborative learning across distributed medical institutions while preserving data privacy.

This work proposes an enhanced LSTM architecture for trend following in time series analysis, improving prediction accuracy and adaptability to various temporal patterns.

This work introduces a surrogate-assisted optimization approach with competitive knowledge transfer mechanisms, enabling efficient search for expensive optimization problems by leveraging historical optimization experiences.

This work proposes an anti-confounding hashing method for radiological image retrieval, utilizing debiased weighting and counterfactual reasoning to eliminate confounding factors and improve retrieval accuracy.

This work develops a dynamic graph representation learning framework for spatio-temporal neuroimaging analysis, capturing both spatial connectivity and temporal dynamics in brain networks.

This work presents a spatial-temporal co-attention learning approach for diagnosing mental disorders from resting-state fMRI data, effectively modeling both spatial and temporal dependencies in brain activity.

This work introduces CausalMixNet, a mixed-attention framework that employs causal intervention techniques for robust medical image diagnosis, addressing confounding factors and improving generalization.

This work proposes an attention-like multimodality fusion framework with data augmentation for diagnosing mental disorders using MRI, effectively integrating multiple imaging modalities and enhancing model robustness.

This work presents an evolutionary sequential transfer optimization approach for solving objective-heterogeneous problems, enabling knowledge transfer across problems with different objective functions.

This work develops a spatio-temporal hybrid attentive graph network for diagnosing mental disorders from fMRI time-series data, combining graph neural networks with attention mechanisms to capture complex brain dynamics.

This work proposes a heterogeneous structured federated learning framework with graph convolutional aggregation for MRI-based mental disorder diagnosis, enabling effective collaboration across distributed medical institutions while preserving data privacy.

This work introduces a dual-stage pseudo-labeling method for diagnosing mental disorders on MRI scans, improving model performance through effective semi-supervised learning strategies.

• Adaptive Text Emotion Recognition Model Training Method, Electronic Device, and Storage Medium
• SVD-Based Heterogeneous Structured Federated Learning for Trend Following of Electromagnetic Radiation Intensity Time-Series
• Multimodal Data Fusion Method, Device, Equipment and Mediums for Speech Recognition
• Subtitle Generation Methods, Electronic Devices, and Storage Mediums Based on Federated Learning
• Disease Diagnosis Methods, Devices, Equipment, and Mediums Based on Multi-Modal Data
• Selection and Trend Following Method of Electromagnetic Radiation Intensity Time-Series

Journal Reviewer

• IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
• IEEE Transactions on Evolutionary Computation (TEVC)
• IEEE Transactions on Medical Imaging (TMI)
• IEEE Transactions on Industrial Informatics (TII)
• Journal of Biomedical and Health Informatics (JBHI)
• IEEE Transactions on Cognitive and Developmental Systems (TCDS)
• IEEE Transactions on Emerging Topics in Computational Intelligence (TETCI)
• Neural Computing and Applications (NCAA)

Conference Reviewer

• International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)
• International Joint Conference on Neural Networks (IJCNN)
• IEEE Conference on Artificial Intelligence (CAI)
• IEEE Big Data