Joanna Zou

Applied Statistics, Stochastics, and ML

My research integrates machine learning models, drawing on data from simulations or experiments, with the governing equations of stochastic dynamical systems to predict complex phenomena in science and engineering.

profile

My projects are on:

Theory

Leveraging tools from probability, inference and information theory, stochastic analysis, optimization, and operator theory to formulate methods with mathematical guarantees.

Algorithms

Developing tractable algorithms for variational inference, sampling, active learning, Bayesian filtering, dimension reduction, and uncertainty quantification, implemented in Python and Julia.

Applications

Bringing theory into practice in AI-for-science disciplines such as molecular dynamics, structural dynamics, diffusion models, optimal control, risk analysis, and digital twins.

About

I am a Postdoctoral Associate at MIT in the Laboratory for Information & Decision Systems (LIDS). In May 2026, I completed my PhD in Computational Science & Engineering with a minor in Analysis & Probability in the Uncertainty Quantification Group at MIT, advised by Youssef Marzouk. During my PhD, I was a visiting fellow with the SFB 1294 Collaborative Research Center for Data Assimilation at the University of Potsdam, hosted by Han Cheng Lie, and a PhD research intern in the Combustion Research Facility at Sandia National Laboratories, hosted by Habib Najm.

Prior to starting my PhD in 2022, I was a Fulbright pre-doctoral researcher at TU Delft hosted by Eliz-Mari Lourens and Alice Cicirello. Previously, I was a software development associate with the NHERI Computational Modeling and Simulation Center and a research intern in the Advanced Technology & Research team at Arup. I received my Masters of Science in Structural Engineering from Stanford University, supported by the Stanford School of Engineering Graduate Fellowship, in 2020. I graduated magna cum laude with a Bachelors of Science in Civil Engineering and minor in Architecture from Columbia University in 2018.

With an interdisciplinary background, I am passionate about fundamental research in mathematics and machine learning as well as applied research in a wide range of fields, including AI-driven materials design, climate forecasting, urban sustainability, renewable energy, and decision-making under uncertainty.

Connect

Email: jjzou (at) mit (dot) edu      LinkedIn      Google Scholar      Github

News

May 2026Our preprint of “Stein kernelized molecular dynamics for active learning of interatomic potentials” is now on ArXiv.
Apr 2026I defended my thesis, “Goal-Oriented Learning of Stochastic Dynamical Systems”, for the PhD in Computational Science & Engineering at MIT.
Mar 2026I gave a talk and co-organized two minisymposia, Advances in MCMC Sampling Methods and UQ for Multiscale Modeling in Computational Chemistry, at the SIAM Conference for Uncertainty Quantification in Minneapolis, MN.
Mar 2026Our preprint of “Goal-oriented learning of stochastic differential equations using error bounds on path-space observables” is now on ArXiv.
Nov 2025I presented a poster on “A goal-oriented loss for learning transition times with machine learning potentials” at the CoMPASs Workshop at the International Center for Mathematical Sciences (ICMS) in Edinburgh, UK.
May 2025I gave a talk and co-organized a minisymposium, Learning Dynamical Systems and Their Observable Statistics, at the SIAM Conference on Applications of Dynamical Systems in Denver, CO, along with Matthew Levine and Han Cheng Lie.
Past Announcements

Current Projects

Robust Stochastic Optimal Control in Reduced DimensionsRobust Stochastic Optimal Control in Reduced Dimensions
Solving stochastic optimal control problems on reduced-dimensional control space
Low-Rank Structure in Drift Functions of SDEsLow-Rank Structure in Drift Functions of SDEs
A spectral method to solve for observable-adapted subspaces of Hilbert spaces of drift functions
Path-Space Approaches to Learning Stochastic Differential EquationsPath-Space Approaches to Learning Stochastic Differential Equations
Adapting error bounds on path-space observables into variational learning objectives for SDEs
Stein Kernelized Dynamics for Active LearningStein Kernelized Dynamics for Active Learning
Adaptively selecting data via particle flows for training energy-based models

Past Projects

Controlled Generative Diffusions for Rare Event SimulationControlled Generative Diffusions for Rare Event Simulation
Adjoint matching and diffusion modeling for sequential approaches to rare event estimation
Training Set Selection by Determinantal Point ProcessesTraining Set Selection by Determinantal Point Processes
A probabilistic approach to D-optimal experimental design
Bayesian Methods for Atomistic Modeling of MaterialsBayesian Methods for Atomistic Modeling of Materials
Multiscale simulation and uncertainty propagation in computational materials science
Numerical Solutions to Fredholm Integral Eigenvalue ProblemsNumerical Solutions to Fredholm Integral Eigenvalue Problems
Nyström and Galerkin projection methods to solving integral eigenvalue problems in KL expansions
Bayesian Filtering for Digital TwinsBayesian Filtering for Digital Twins
Robust structural health monitoring via Gaussian process latent force models
Heteroscedastic Gaussian Process Surrogate ModelingHeteroscedastic Gaussian Process Surrogate Modeling
Surrogate modeling for regional-level seismic simulation
High-Performance Computing for Probabilistic Hazard AnalysisHigh-Performance Computing for Probabilistic Hazard Analysis
Risk-informed decision making for natural hazards
Image Segmentation for Land Use ClassificationImage Segmentation for Land Use Classification
U-Net classifier of early crop types from satellite imagery