Optimal Online Date Driven Optimization with Multiple Time Varying Non Convex Objectives
Abstract
The main objectives of the proposed activities are design of : [1] a real time prediction framework for modeling and predicting from heterogeneous data with multiple objective function, [2] a system that detects change points online, builds dictionary for recurring patterns, and predicts rare events of commander’s interest, [3] a framework that adaptively optimizes the commander’s decision by successively combining machine’s analytic results from different data sources and/or objective functions, [4] a framework to learn from and predict commander’s actions and objectives, and to efficiently incorporate commanders inputs/prior knowledge by data-driven optimization and control, [5] a modeling method to facilitate commander’s evolving (time-varying) objectives that may not be exactly known (quantitatively), and/or may have potential trade offs with each other, and [6] an online interactive optimization to improve deep learning by reducing the amount of required data (whose statistics may be time varying) during training and testing process.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 29, 2019
- Source ID
- FA86501817837
Entities
People
- Vahid Tarokh
Organizations
- Air Force Research Laboratory
- Defense Advanced Research Projects Agency
- Duke University