Gorodetsky Group

We build methods that enable prediction, inference, optimization, and decision-making in settings that are too expensive, too uncertain, or too high-dimensional for standard approaches, creating new capabilities for scientific discovery, engineering design, and complex system operation. Our research centers on multi-fidelity uncertainty quantification, inference, and optimization; Bayesian inference and learning in static and dynamic systems; and tensor methods for scalable scientific computation.

These methods address scientific and engineering systems where data are limited, information is indirect, models are imperfect, and uncertainty is unavoidable. They are foundational to the predictive, adaptive, and uncertainty-aware computational workflows now emerging in digital twins, scientific machine learning, and AI for science. We develop and deploy ideas in a variety of applications, including electric propulsion, fluid mechanics, radiation and plasma transport, and aerospace sensing, estimation, and control.

Driving questions

How do we make useful predictions from limited, indirect, and heterogeneous information?
How do we allocate resources to data collection, modeling, and computation intelligently under uncertainty?
How do we build scalable algorithms for high-dimensional and multiscale physical systems?
How do we enable reliable design and decision-making when first-principles understanding alone is not enough?

Research ecosystem

Collaborations and implementations. This work is developed through collaborations with the National Aeronautics and Space Administration (NASA), Sandia National Laboratories, and Los Alamos National Laboratory. Related methods are implemented in scientific-computing ecosystems including the U.S. Department of Energy's Dakota toolkit, PyApprox, and NASA's MXMCPy.

University and multi-institution programs. The group participates in the Michigan Institute for Computational Discovery and Engineering (MICDE), the Michigan Institute for Data and AI in Society (MIDAS), the Plasmadynamics and Electric Propulsion Laboratory (PEPL), and the Center for Computational Medicine and Bioinformatics (CCMB), and contributes to the Joint Advanced Propulsion Institute (JANUS), Space Power and Propulsion for Agility, Responsiveness, and Resilience (SPAR), the Michigan-Los Alamos Strategic Partnership and Accelerating Research Collaboration (MI-SPARC), and the Center for Prediction, Reasoning & Intelligence for Multiphysics Exploration (C-PRIME).

Research support. Support from the National Science Foundation (NSF), the Air Force Office of Scientific Research (AFOSR), the Department of Energy (DOE), the Defense Advanced Research Projects Agency (DARPA), NASA, and Draper sustains foundational algorithm development and application-driven work.

Tensor-network radiation transport relevant to fusion plasma modeling

How computing and AI could help unlock fusion energy

SPARC x U-M / Los Alamos

In the SPARC collaboration with Los Alamos, we are developing accelerated radiation transport methods for fusion-relevant plasma modeling. The broader effort connects high-performance computing, AI, plasma physics, and scalable numerical methods, with reported speedups of 100 times or more for realistic-scale problems.

Read the U-M / LANL feature

How computing and AI could help unlock fusion energy

1 / 4

Applications

We apply these methods in electric propulsion, fluid mechanics, radiation and plasma transport, and aerospace problems involving sensing, estimation, and control. These domains help us develop, test, and refine the research.

Electric propulsion: Hall thruster uncertainty quantification, simulation acceleration, experimental design, and predictive modeling in collaboration with propulsion researchers and NASA-linked programs.
Fluid mechanics: Multi-fidelity inference, uncertainty quantification, and optimization for flow problems where high-fidelity computation is expensive and uncertainty matters materially.
Radiation and plasma transport: Tensor-based methods for transport and predictive plasma computation, including recent work with Los Alamos collaborators on thermal radiation transport.
Aerospace sensing, estimation, and control: Inference, uncertainty-aware learning, and optimization for sensing, trajectory, and decision problems in aerospace systems.