Projects and training

Expectation propagation in linear regression models

Research training, Heriot-Watt University, School of Engineering and Physical Sciences, from 2021-09 to

Learn the basics of variational methods, including VB (MFVB and FFVB) and EP (parallel EP, AEP, SEP and α-EP) methods. Compare different variational methods in linear regression model. Explore the algorithm design with different problem-specific convariance constraints.

Review of “New error bounds for deep ReLU networks using sparse grids”

Course project, University of Edinburgh, School of Mathematics, 2021-03

This is a paper aims to address the problem, why and when deep networks can lessen or break the curse of dimensionality. Instead of focusing on a particular set of functions which have a very special structure, they consider functions in the Korobov spaces which is more general for high dimensional multivariate approximation.

Regression models in JAGS

Course project, University of Edinburgh, School of Mathematics, from 2020-11 to 2020-12

Compare three Bayesian models by implementing the models in JAGS, having JAGS sample from the corresponding posterior densities, and then using the deviance information criterion, DIC, for model comparison.

Bernstein type problem for minimal hypersurfaces

Undergraduate dissertation, Beihang University, School of Mathematical Science, from 2020-02 to 2020-05

Introduce the value distribution problem of Gauss map of minimal surfaces in Euclidean space and present major contributions by Osserman,Xavier and Fujimoto.

Rational solutions to an extended KP-like equation with symbolic computation

Course project, Beihang University, School of Mathematical Science, from 2019-10 to 2019-12

An introduction to the Hirota bilinear method. Propose an extended KP-like equation with the generalized bilinear operators and discuss rational solutions based on polynomial solutions to the generalized bilinear equation.