My research area lies in data science. I primarily study and develop machine learning tools based on linear algebraic techniques, including those that allow for transparency and efficiency in large-scale problems. Under this framework, I address various problems and applications in data science.

I am always open to discussing research. Please feel free to email me. Below are brief descriptions of some problems and applications I have worked on. You can also find my publications on Google Scholar.

Fairness and Bias in Machine Learning

Machine learning, particularly automated decision-making, is increasingly used in sensitive domains, yet critical issues such as gender, racial, and social bias remain insufficiently addressed.  I have a general interest in studying algorithmic fairness and developing transparent machine learning models that mitigate bias from both data and algorithms.

Topic Modeling

Topic modeling is an unsupervised machine learning technique used to reveal hidden patterns in large datasets. A popular approach for topic modeling that provides a low-rank approximation of a matrix is nonnegative matrix factorization (NMF). My work investigates NMF and its variants, including supervised and fairness-aware formulations, for different applications

Tensor-based Techniques

Tensors, multidimensional generalizations of matrices, enable the analysis of complex, multi-modal data. However, compared to matrices, there are fewer well-established mathematical techniques for tensors due to their theoretical and practical challenges. My work develops and applies tensor-based techniques, such as tensor decompositions for topic modeling and tensor-based iterative methods for solving tensor linear systems.

Data Completion Techniques

Matrix completion is the task of imputing missing entries of a partially observed matrix. Data completion is essential, whether as the main goal in recommender systems or as a pre-processing step for learning tasks. I have a general interest in matrix completion techniques and downstream tasks with incomplete data. My prior work presents an iterative method for low-rank matrix completion that incorporates sparsity-based structure in the missing entries.

Dimensionality Reduction Techniques

Dimensionality reduction is the task of reducing the number of features in a dataset while best preserving certain aspects of the data. I have a general interest in dimensionality reduction techniques and studying their properties. My prior work seeks to understand the classical Multidimensional Scaling algorithm, a popular technique for pattern recognition and data visualization. I am also interested in fairer formulations for dimensionality reduction techniques.