My Projects
Disentangled Representation Learning with Less Supervision

Proposed a novel ControlVAE that combines control theory with Variational Autoencoders to disentangle the latent factors.
 Apply ControlVAE to text generation in Ecommerce platform
Physcis Guided Machine Learning

Incorporated the laws of physics into deep learning models to improve their interpretability and performance

Apply the proposed models to COVID pandemic, autonomous cars and climate changes
Data Analysis in Social Networks

Developed a novel optimization method of choosing right sources in social networks to solicit data from to minimize the expected error given a limited cost budget

Adopted the Expectation Maximization (EM) algorithm to estimate the reliability of sources on Twitter

Formulated a zeroone integer nonlinear programming (INLP) problem and then proposed a similaritybased lossy estimation algorithm to solve the modeling effectively