Autonomous Learning in a Dynamic World
Abstract
This project will develop new machine learning algorithms that adapt autonomously to change. This capacity is critical, as change is constant whereas it is rare that the system modeled remainsstatic. In contrast, most machine learning research focuses on learning static models. Such models inevitably fail to be as relevant to current circumstances as they were to those in which they were learned. Those algorithms that do adjust to drift assume that drift is uniform, whereas in practice it is likely to differ in form and rate in different data subspaces. For example, the way in which men s purchasing behavior changes may differ from the way in which women sbehavior changes and changes in the behavior of the young may differ from changes in the behavior of the old. This project will develop new techniques that can adapt autonomously and dynamically to different types and rates of drift in different parts of the distribution from which the stream of data is sampled. They will deliver effective autonomous capacity for continuous learning in a dynamic world.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2017
- Source ID
- FA23861714033
Entities
People
- Geoff Webb
Organizations
- Air Force Office of Scientific Research
- Monash University
- United States Air Force