Visiting Research Student at University of Michigan, ILLIDAN Lab (2024-01 – Present)
- Clinical ML. Led two first-author clinical AI studies developing a cross-cohort surrogate-biomarker framework that harmonized shared clinical variables across NACC and I-CONECT and linked conversational language with multi-modality biomarkers.
- Built the end-to-end modeling pipeline, including data preprocessing and harmonization, training and evaluating established machine-learning models, and Optuna-based hyperparameter optimization.
- Statistical validation through feature-importance, correlation, and sensitivity analyses.
- Foundations of LLMs and sequence modeling.
Developing
Fourier-based sequence-modeling blocks as computationally efficient alternatives to attention for long-context and irregularly timed sequences, with a focus on multiscale temporal dependence, causal modeling, and scalable architectures; benchmarking the designs on synthetic sequences and EHR next-event prediction.
Graduate Research Assistant at Michigan State University (2021-01 – Present)
- Adam theory. Established the first convergence guarantees for plain vector-formulated Adam under heavy-tailed stochastic noise through a generalized discounted-to-nonconvex conversion and discounted-regret analysis.
- Zeroth-order optimization. Proposed forward-pass-only methods for deep-model optimization; proved optimal query complexity and achieved substantial improvements across diverse vision and language tasks.
- Clinical ML. Led the full lifecycle of a first-author longitudinal study predicting progression from normal cognition to mild cognitive impairment/Alzheimer's disease, including raw-data preprocessing, cohort construction, feature selection, model development, and rigorous evaluation; Mentored and technically led an undergraduate researcher through three projects on subject harmonization, distributed batch-effect correction, and bilingual speech-based cognitive assessment, resulting in publications at PSB, KDD, and Interspeech.
