Context-aware learning- Towards intelligent decision-making in science and engineering
Abstract
Data-driven decision-making in science and engineering typically is a two-step process. The first step learns a model of the underlying system from data and the second step uses the learned model to solve an outer-loop task such as optimization, uncertainty quantification, and control for informing upstream decision makers. However, splitting data-driven decision-making into two steps means that a general-purpose model is learned in the first step that is agnostic to the typically rich context given by the outer-loop task, user tolerances, available resources and other properties of the specific setup of the second step. In stark contrast, this project introduces the concept of context-aware learning that combines, instead of separates, data-driven modeling and solving outer-loop tasks. By learning a model for the specific task and context at hand, the proposed contextaware learning methods more efficiently balance budgets, collect data, and train models to enable data-driven decision-making with much relaxed requirements on number of data points, data quality, and training costs compared to traditional methods. Taking into account the context and profiting from the relaxed requirements of context-aware learning is particularly beneficial for decision-making based on systems with nonlinear dynamics, high-dimensional states, and many uncertain parameters, which makes learning accurate general-purpose models challenging and often impossible in practice where data are scarce and training time is limited. Applications in uncertainty quantification and robust control demonstrate that context-aware learning has the potential to impact a wide range of decision-making problems with complex engineering systems and limited resources with relevance to the Air Force mission.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502110222
Entities
People
- Benjamin Peherstorfer
Organizations
- Air Force Office of Scientific Research
- New York University
- United States Air Force