cv
Summary
Name | Hyemin Gu |
Research interest | generative modeling, dynamical transport, gradient flows, particle transport, Wasserstein proximal regularization, entropic regularization |
Languages | English, Korean |
Contact | hgu@umass.edu |
Education
-
2020 - Present PhD candidate in Mathematics
University of Massachusetts Amherst, MA, US - Specializing in Mathematical machine learning
- Advised by Markos Katsoulakis
-
2018 - 2020 MSc in Mathematics
Ewha Womans University, Seoul, Republic of Korea - Thesis "Convolutional Neural Network for 2D Flow Estimation Problem"
- Specializing in Numerical analysis
- Advised by June-Yub Lee
-
2014 - 2018 BSc in Mathematics
Ewha Womans University, Seoul, Republic of Korea - Second major in Computational Science and its thesis "Low cost training of a classification Neural Network with respect to Weight Selection"
- Minor in Statistics
- Dean’s list 5 semesters
Academic Interests
- dynamical transport, gradient flows, particle transport
- Wasserstein proximal regularization, entropic regularization
- generative modeling
Conferences
-
2023 Poster presentation
Optimal Transport in Data Science, ICERM, Brown university - H. Gu et al., Lipschitz Regularized Gradient Flows and Latent Generative Particles
-
2018 Poster presentation
Joint Mathematics Meetings, Mathematical Association of America - H. Gu, Training a 2 layer Neural Network using SVD-generated weights
-
2017 Poster presentation
Joint Mathematics Meetings, Mathematical Association of America - J. Park et al., Necessary and sufficient conditions for shortest vectors in lattices of dimension 2 and 3
Projects
-
Lipschitz regularized generative particles algorithm
- Implementation of a particle transportation algorithm through gradient flows associated with Lipschitz regularized f-divergences.
- A generative model alternative to GANs in low training data regimes.
- Mathematical interpretation of applying spectral normalization on neural network discriminators as a particle transportation speed regularization.
-
Wasserstein-1/Wasserstein-2 proximal generative flow
- Formulated and implemented a generative model with continuous-time adversarial flow architecture for learning distributions that are supported on low-dimensional manifolds.
- Our formulation is analyzed via Mean Field Game theory and ensures good properties of the learned flow such as uniqueness of solution and optimal (linear) paths.
Training
-
2018 Industrial Mathematics Academy
National Institute for Mathematical Sciences, South Korea - Presented a final result of a group project for solving industrial problem
- Proposed a Convolutional Neural Network for classifying infected individual from images
- Coordinated team efforts for the group project
- Attended tutorials on Python data analysis and Keras, lectures on matrix based data analysis, linear programming theory and practice
-
2017 Industrial Mathematics Academy
National Institute for Mathematical Sciences, South Korea - Proposed a model for assessing safe driving scores from Onboard diagnostic data based on Poisson process
- Attended tutorials on basics to neural networks
Teaching Experience
-
2021 Graduate teaching assistant
University of Massachusetts - Amherst, Amherst, MA, US - (MATH532H) Nonlinear dynamics and chaos with applications; graded assignments and conducted tutorial sessions for Python ODE solving
- (MATH545) Linear algebra for applied mathematics; graded assignments, conducted discussion sessions and arranged offie hours
- (MATH235) Linear algebra, (MATH545) Linear algebra for applied mathematics; graded assignments
-
2018 - 2019 Graduate teaching assistant
Ewha Womans University, Seoul, Korea - Numerical differential equations (numerics for ODE/PDE, Monte-Carlo, optimization); graded assignments and arranged office hours
- Finite mathematics and programming (Matlab programming, mathematical logic, combinatorics); graded assignments and arranged office hours
- Calculus 2 (multivariate calculus); graded assignments and arranged office hours
- Numerical analysis (linear system solving, power method, numerical integration/differentiation); graded assignments and arranged office hours
Work Experience
-
2020 Statistics specialist
Ewha Womans University Seoul Hospital, Seoul, Korea - Developed a pipeline for acquiring, analyzing, and visualizing gene expression data from open repositories using R; authored a tutorial book on the process
- Conducted training sessions on statistical analysis using R for colleagues
Leadership
-
2022 - Present Department seminar organizer
Department of Mathematics and Statistics, University of Massachusetts Amherst, MA, USA - Invited speakers among faculties in the department for introducing their research interests to early career graduate students and hosted talks
- Participated in regular Graduate Student Advisory Committees meetings, reported the progress, and discussed future directions
Honors and Awards
-
2024 Mathematics and statistics department in University of Massachusetts Amherst - Anne and Peter Costa Graduate Prize in Applied Mathematics