Hi, I am Teresa

I am a PhD candidate in the Department of Applied Mathematics & Statistics at Johns Hopkins University. I am very fortunate to be co-advised by Professor Soledad Villar and Professor Carey Priebe.

I have no special talent. I am only passionately curious. In particular, I am passionate about understanding learning — including machine learning and natural learning — from mathematical and statistical principles.

My research interests are in the areas of representation learning and deep learning. My current work focuses on expressivity and generalization properties of graph neural networks.

Interests

  • Representation Learning
  • Graph Neural Networks
  • Interdisciplinary Research in Data Science

Education

  • Ph.D. in Applied Math & Stat, 2024

    Johns Hopkins University

  • M.S. in Data Science, 2020

    New York University

  • B.S. in Statistics and Economics, 2016

    University of Hong Kong

Research

Approximately Equivariant Graph Networks

We focus on the active symmetries of GNNs, and show a bias-variance tradeoff controlled by the choice of symmetry.
PDF

Fine-grained Expressivity of Graph Neural Networks

We quantify which distances MPNNs induce, leading to a fine-grained understanding of their expressivity.
PDF

From Local to Global: Spectral-Inspired Graph Neural Networks

We propose spectral-inspired GNNs that exploit the advantages from both global and local methods.
PDF

Deep Learning is Provably Robust to Symmetric Label Noise

L1-consistent DNNs can tolerate massive symmetric label noise up to the information-theoretic threshold.
PDF

Dimensionality reduction, regularization, and generalization in overparameterized regressions

We show PCA avoids the peaking phenomenon of double-descent, and overparameterization may not be necessary for good generalization.
PDF Poster
See all publications

Experience

 
 
 
 
 

AI/ML Research Intern

Apple ML Research

May 2023 – Present Cambridge, UK
Developing approximately equivariant diffusion models.
 
 
 
 
 

Research Associate

Center for Computational Mathematics, Flatiron Institute

Jun 2022 – Aug 2022 New York City, USA
Studied expressivity of graph neural networks.

Projects

*
All Representation Learning Interdisciplinary Research Data Visualizations

Autonomous driving

A CNN encoder-decoder model to navigate traffic environment using bird's eye view images.

Topic Modelling with Latent Dirichlet Allocation (LDA) on Wikipedia Articles

An interative visualization showing results of topic modelling on 33k Wikipedia articles

Chatbot

An interactive jupyter notebook implementing live chatbot using N-gram blocking

Matrix Completion for Different Missing Data Patterns

A study of three matrix imputation methods on different types of missing data patterns

Water Conservation

A data visualization showing global inequities of water usage and conservation suggestions

How Crypto Stacks Up Against Other Investments

A data-driven graphic story. Collaboration with Yue Qiu, Justina Lee and Adrian Leung in Bloomberg.

Popular Topics

Deep Learning Generalization Graph Neural Networks Interdisciplinary Research Label noise Representation Learning Theory of deep learning Visz