Learning from Dynamics on Metric Spaces and Graphs
Abstract
The PI, Dr. Mauro Maggioni, proposes to research how to exploit observations of time-varying data to construct (i) statistical models for the data and its dynamics, (ii) statistically and computational efficient estimators for the parameters of such models, and (iii) re-use information acquired by observing one system to perform similar statistical learning tasks more quickly on new, possibly more complex systems. Especially, he is interested when the underlying geometry of data is non-Euclidean. A key motivation arises from emphasizing the importance of the time-varying aspect of many data sets in a wide variety of settings, from learning efficient representations - in time and space - of shapes and objects in a visual scene, to exploiting many short trajectories simulated in parallel of a high-dimensional stochastic system to perform model reduction, or discover laws in complex systems of interacting agents.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 06, 2024
- Source ID
- FA95502310445
Entities
People
- Mauro Maggioni
Organizations
- Air Force Office of Scientific Research
- Johns Hopkins University
- United States Air Force