Autonomous Action by Learning Group Action Protocols and Case-Based Reasoning
Abstract
This project challenges a difficult machine learning problem that learns a sequence of actions from big heterogeneous data of mixed types. The problem is to infer appropriate actions for a new object, given a large set of heterogeneous and temporal objects, each associated with temporal actions that follow an autonomous action protocol. Three new bodies of theory and techniques are to be developed to solve this problem: 1) clustering heterogeneous and temporal objects into subgroups of similar objects, 2) learning an autonomous action protocol for each subgroup, and 3) recommending actions for a new object in its action process based on the learned protocol of the subgroup it belongs to and the actions of its nearest neighbors. The data this project uses is Electronic Medical Record, which is best suited to this problem, but the framework is general andcan be applied in different autonomous systems with appropriate adaptations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 09, 2018
- Source ID
- FA23861714094
Entities
People
- Tu-bao Ho
Organizations
- Air Force Office of Scientific Research
- John von Neumann Institute for Computing
- United States Air Force