Model Reduction Group

Home

Overview

The Model Reduction Group (MRG) at MIT is a conglomeration of graduate students, postdoctoral associates, and professors from a wide variety of research groups, all united by an interest in model reduction. Our members come from the Aerospace Computational Design Laboratory (ACDL), Computational Prototyping Group (CPG), Computer Science and Artificial Intelligence Laboratory (CSAIL), Laboratory for Electromagnetic and Electronic Systems (LEES), and the Combustion Dynamics Laboratory (CDL). Members include students, postdocs, and professors from Electrical Engineering and Computer Science, Aeronautics and Astronautics, Mechanical Engineering, and Chemical Engineering. The group meets weekly in Room 33-206 on Monday at 2pm. Meetings are intentionally informal; we hope to stimulate discussion between active researchers in all extents of model reduction algorithms and applications. The focus of the group is on existing challenges in model reduction -- both for the development of new algorithms (e.g., sampling techniques) and for the extension to new applications (e.g., chemical kinetics).

The purpose of MRG is to promote cross-fertilization of ideas, concepts, and challenges among researchers at all levels with varying backgrounds and future interests. We strive for two concrete research goals: (1) to match new applications with (the extension of) appropriate methodologies and (2) understand and attack the existing challenges for model reduction algorithms.

Maintaining interpretability in reduced models
Managing widely-varying empirical parameterization in large-scale compartmental models
A posteriori error analysis for reduced models
Choosing the basis functions in the projection framework
Model reduction algorithms for nonlinear equations
Deriving/using reduced models for optimal control problems
Time-varying basis
Reduced model treatment of uncertainty in statistical setting
System identification methods
Formulating model reduction problems as efficient optimization problems
Systems with large number of parameters
Benchmark examples
Overview of prepackaged (such as those in MATLAB) routines
Interconnecting reduced models back into larger systems