cv

Research Intern

Orby AI - A Uniphore Company

Studying preference learning for LLM fine-tuning, honored to be advised by Peng Qi and Gang Li.

Ph.D.

Georgia Institute of Technology

Statistics

Bachelor

University of Science and Technology of China

Statistics

Runner Up

Alice and John Jarvis, Ph.D. Student Paper Competition

Mentoring Service Award

SURE program, Georgia Institute of Technology

Stewart Fellowship

Georgia Institute of Technology

Annual outstanding Student Scholarship (2017-2020)

University of Science and Technology of China

Kernel-based Equalized Odds: A Quantification of Accuracy Fairness Trade-off in Fair Representation Learning

The Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025)

It provides a quantification for the trade-offs between three categories of fairness notions, including independence, separation, and calibration, delivering an equalized-based fairness metric that helps achieve fairness for a class of downstream tasks.

Abductive Preference Learning

arXiv

It proposes a solution for the over-confidence phenomenon for SOTA LLM models, serving as a fine-tuning method that detects the modification of keywords in text prompts or items in image prompts.

A Uniform Concentration Inequality for Kernel-Based Two-Sample Statistics

Reject and Resubmit with Journal of Machine Learning Research

It provides a general inequality to establish finite sample bounds for optimization problems where kernel-based statistics are involved as part of the objective function, revealing that the convergence rate or error bound for kernel-based statistics is relevant with the input dimension only up to a logarithm factor.

Sequential and Simultaneous Distance-based Dimension Reduction

Journal of Nonparametric Statistics

This work studies a stepwise dimension reduction method, built without the elliptical distribution assumption for widely studied Sufficient Dimension Reduction (Method), detecting nonlinear subspaces for both response and predictors.

Simultaneous Dimension Reduction and Variable Selection for Multinomial Logistic Regression

INFORMS Journal on Computing