Welcome to NJIT Xu Lab!

I am an Assistant Professor in the Department of Data Science, Ying Wu College of Computing at NJIT. I also hold a Research Affiliate position at MIT, associated with the NSF Center for Brains, Minds and Machines (CBMM) at McGovern Institute for Brain Research. Before joining NJIT, I worked as an Assistant Professor (Research) in the Division of Applied Mathematics at Brown University collaborating with Prof. George Em Karniadakis on stochastic graph embedding methods for static/temporal graph representation learning, and also working with Prof. Tomaso Poggio on machine learning theory including efficient large language model optimization and generalization, as a joint postdoc at McGovern Institute for Brain Research at MIT. Moreover, I collaborate with Dr. Dimitrios Pantazis from the MEG lab (part of the Athinoula A. Martinos Imaging Center) at MIT and Prof. Quanzheng Li from the Massachusetts General Hospital, Harvard Medical School on functional neuroimaging data analysis for Alzheimer's disease early-stage prediction and cognitive training effect evaluation. I obtained my PhD degree from the Department of Computer Science, Northeastern University in Shenyang, China advised by Prof. Hong Zhao (Co-founder of the Neusoft Corporation) and spent two years as a joint PhD student in the Division of Applied Mathematics, Brown University supervised by Prof. George Em Karnidakis. During my PhD, I spent the first two years working at Neusoft Research Institute on developing effective medical image analysis tools for human brain MRI imaging data analysis.

Research Interests

Our research lies at the intersection of Computer Science, Medical Imaging, and Neuroscience. The main goal is to develop explainable, accurate, and efficient deep learning approaches for diverse practical applications.

• Machine Learning Theory: Dynamics in deep neworks with square loss (Research2023), Norm-based Generalization Bound (NeurIPS2023), SGD vs. GD: Asymptotic noise and Rank Shrinkage (Preprint2023), and Generalization of Deep Networks (CBMM Memo112, 2020).
• Spatio-Temporal Graph Representation Learning with Uncertainty Quantification for Computational Neuroscience (i.e., fMRI/MEG brain connectivity): Healthy Brain Agining Trajectory Detection with MEG Brain Networks (Preprint, 2024); Hyperbolic MEG Brain Network Learning for Subjective Cognitive Decline Prediction (IEEE JBHI 2024). Alzheimer's Disease Progression Prediction with MEG (TBME2021); Assessment of Non-Pharmacological Cognitive Traning Effects with fMRI (PLoSComp.Bio.2020); Two recently developed dynamic graph represenation learning methods based on transformers (Neural Networks2024, TNNLS2022).
• AI for Healthcare: Diabetics retinal microaneurysm detection with AOSLO-net (Translational Vision Science & Technology,2022); Automated Red Blood Cell Segmentation and Classification for Sickle Cell Anemia via deformable U-Net (JBHI2020, MICCAI2018) and via deep CNNs (PLoSComp.Bio. 2017).

Furthermore, we are also working on developing graph diffusion models and causal inference for climate modeling supported by the DOE grant (Co-PI, 2023 - 2027).