The PI's primary research interests are in the fields of medical physics, big data mining, machine learning, deep learning, and bioinformatics. His research focuses on the development, analysis, implementation, and experimental evaluation of big streaming data mining algorithms, machine learning and deep learning techniques, and applications in healthcare and medical physics.

LOAD Lab's long-term research interest lies in advancing the capabilities of current online learning and deep learning techniques for big data streams and healthcare data mining. We are highly interested in optimization theory, deep learning, and network integration techniques for real-world applications. At the beginning of the research career at Xavier, the PI is mainly on developing machine learning and data mining models for tackling unsolved practical challenges in big data streams with various application domains, including but not limited to, fraud/intrusion detection, novel class detection, cancer subtype and state prediction, and cancer disparities. The PI is also interested in developing abnormal detection and novelty detection methods to depict unexpected and novel classes from massive streaming data, which are based on online learning techniques including CSTG, CSRDA, ALoKDE, CSGOL, SPDA, OSLMF, RSOL, ASOL, and ATG. He has also investigated many machine learning and deep learning methods with the applications to healthcare domains, such as Fusion Lasso, Dropfeature-DNNs for cancer subtype and stage prediction, and deep latent variable models for inferring personalized and race-specific causal effects.

Specific research areas at LOAD Lab include:

Model Interpretability: Studying the interpretability of sparse online learning and deep learning models.

Online Learning and Optimization: Learning incremental models continuously from big streaming data.

Feature Evolvable Online Learning: Learning effective models from the data streams with features that would evolve in the open and dynamic environment.

Semi-supervised Learning: Learning both from labeled and unlabeled (noisy) data.

Deep Learning and Applications: Leveraging deep neural networks to handle complex multi-omics genomic data and better cancer radiation therapy.