Fast and memory-efficient method to sequentially sample knockoffs
"The knockoff filter is a general framework for controlling the false discovery rate when performing variable selection" - Prof. Candes.
I worked on building an algorithm that efficiently sample knockoff sequentially that controls FDR without losing statistical power under the supervision of Prof. He. The algorithm is implemented in both R and Python.

Conditionally simulating high-resolution galaxy images by deep generative models
Our aim in this work is to conditionally simulate high-quality images for specific galaxy populations by taking physical properties as input. We explore the use of deep generative models to achieve our goal. We propose a novel two-step CVAE-SRGAN method that conditions on several physical properties and samples images from the latent distribution. Our results suggest that our approach not only generates very realistic galaxy images with high resolution and complex structures, but achieves a very good prediction accuracy and conditional effect as well.

Computing Wasserstein Barycenter
We describe and implement several interior point algorithms for linear programming, specifically on the computation of the Wasserstein barycenter. The questions focus on both the case of Pre-specified Support Problem and on the general Free Support Problem using different interior-point methods. We also compare the computational efficiency of those methods.