He (Evelyn) Lyu

PhD in Computational Mathematics, Science, and Engineering

image

I got my PhD in Computational Mathematics, Science, and Engineering from Michigan State University in summer 2023, supervised by Dr. Rongrong Wang. Before entering MSU, I obtained my bachelor degree in Mathematics and Applied Mathematics from Fudan University.

My research interests include Machine learning, Compressed sensing, Statistical analysis, and Optimization.


News

  • Nov 27th 2023 - I started a new journey at Meta as a Research Scientist in Ads Core ML team!
  • June 26th 2023 - I am excited to announce that I joined AADS team at Eli Lilly and Company!
  • May 21st 2023 - Our paper Quantization of Bandlimited Graph Signals was published at SampTA 2023! [link]
  • May 19th 2023 - I successfully defended my PhD dissertation! My gratitude goes to my advisor Dr. Rongrong Wang and my committee members.
  • Jan 2023 - Our paper Sigma Delta Quantization for Images was published at Communications on Pure and Applied Mathematics! [link]
  • Jan 2023 - Our paper Implicit regularization in Heavy-ball momentum accelerated stochastic gradient descent was accepted as a spotlight paper at ICLR 2023! [link]
  • Aug 2022 - I have completed my summer internship at Meta as a ML Software Engineer intern.
  • Sep 2019 - Our paper Manifold denoising by nonlinear robust principal component analysis was accepted at NeurIPS 2019!
  • Aug 2018 - Started a new journey at MSU!

Publication

* denotes equal contribution
  • Ghosh, A.*, Lyu, H.*, Zhang, X. and Wang, R., 2023. Implicit regularization in Heavy-ball momentum accelerated stochastic gradient descent. Spotlight paper at ICLR 2023. [pdf]
  • Lyu, H. and Wang, R., 2020. Sigma Delta quantization for images. arXiv preprint arXiv:2005.08487. Communications on Pure and Applied Mathematics. [pdf]
  • Lyu, H., Sha, N., Qin, S., Yan, M., Xie, Y. and Wang, R., 2019. Manifold denoising by nonlinear robust principal component analysis. NeurIPS 2019. [pdf] [MATLAB code] [Python code]
  • Lyu, H. and Wang, R., 2020. An exact sin Theta formula for matrix perturbation analysis and its applications. arXiv preprint arXiv:2011.07669. Under review. [pdf]
  • Lyu, H. and Wang, R., 2022. Perturbation of invariant subspaces for ill-conditioned eigensystem. arXiv preprint arXiv:2203.00068. [pdf]

Industrial Experiences

Machine Learning Software Engineer Intern

Meta | May 2022 - Aug 2022
  • Worked on discovering effective Machine Learning design principles to optimize Ads Ranking models.
  • Designed and implemented experiments to test the effectiveness of seven ML modeling techniques on Ads Ranking models, derived a statistical method to measure the effectiveness quantitatively.
  • Identified effective modeling techniques and refined the search spaces for seven Ads Ranking models with total revenue share 4.12%. Obtained NE (most important metric for measuring model performance) gain of over 0.5% on 4 models.

Research Experiences

Implicit Regularization in Heavy-ball Momentum SGD

MSU | May 2022 - Sep 2022
  • Derived an implicit regularization analysis for Stochastic Gradient Descent with Heavy-ball momentum, which provides theoretical insights on how momentum affects the generalization performance of SGD.
  • Validated the theoretical analysis by numerical experiments on image classification tasks using large Convolutional Neural Networks and real-world datasets including Computer Vision datasets CIFAR10 and CIFAR100.

Sigma Delta Quantization for Images

MSU | Sep 2018 - May 2020
  • Proposed and analyzed an adaptive quantization method for direct digital image acquisition that obtains a better information conversion rate than the state-of-the-art method in cameras.
  • Designed and implemented a scalable algorithm for solving the optimization problem involved.
  • Patent pending.

Matrix Perturbation Analysis and Its Statistical Applications

MSU | Jun 2020 - Dec 2020
  • Established a set of a collection of improved error bounds on SVD perturbation related problems.
  • The improved error bounds can be applied to clustering, classification, and dimension reduction methods.

Manifold Denoising by Nonlinear Robust PCA

MSU | Mar 2019 - May 2019
  • Proposed and analyzed an algorithm that extends robust principal component analysis (RPCA) to nonlinear manifolds, which can be applied to manifold denoising tasks.
  • Applied FISTA algorithm to solve the optimization problem involved.

More about me

  • I'm a Amateur 5-dan Go(Weiqi) player, and I've been learning Go since I was 7. Sedol Lee has always been my favorite Go player. For my other hobbies, I enjoy reading history books and visiting museums.