layout
default

About me

Prior to USC, I was an undergrad in mathematics at Central University of Finance and Economics.

My reseach interest lies in High-dimensional inference and Statistical/interpretable machine learning. I am currently working on:

A regularization path i.e, LASSO, SCAD, Elastic Net solution path and etc. related method to provide p-values for high-dimensional models.
A unified framework for calculating variable importance/feature importance for machine learning algorithms involving regularization.

Publications

Cao, X, Gregory, K.B, Wang, D, A generalized framework for high-dimensional inference using Leave-One-Covariate-Out regularization path, ready to submit.
- Our procedure allows for calculating variable importance, variable screening/selection and statistical inference. Test statistics constructed by calculating the change in LASSO solution path. P-values are estimated by bootstrapping the null distribution.
- We may outperform the state-of-the-art.
- R package LOCOpath available.
- A quick demo of our method. This demo is coded in Python.
Cao, X, Gregory, K.B, Wang, D, Leave-One-Covariate-Out regularization path on Generalized Linear Models, Cox models and Gaussian Graphical Models, manuscript in preparation.
- An extension of LOCO regularization path to GLM, cox PH and graphical models.
Cao, X, Gregory, K.B, Generalized L1 regularization for Mixture Regression Models, manuscript in preparation.
- Use generalized LASSO penalty to adaptively control different components of mixture.