Multi-Source Learning Coupled with Reasoning Capabilities for Large-Scale Decision Support Environments
Abstract
Machine learning paired with reasoning mechanisms provides the core of many modern intelligent systems, such as cyber-physical systems and smart transportation systems. However, reasoning with knowledge and information extracted from multiple data sources in these large-scale systems still remains a significant challenge, due to the problem of data heterogeneity and uncertainty. The problems are (1) traditional data integration approaches would have to transform data into a common representation, potentially losing the domain specific knowledge along the way, and (2) integrating information after making decision/learning for each data source is optimal for each source but the hard-coded knowledge has dropped valuable information for integration with other sources. The aim of this research is to investigate these challenging issues in such distributed decision support environments so as to develop a novel multi-agent based methodology for learning, reasoning and fusion of knowledge discovered from heterogeneous sources of data and domain knowledge, which is capable of handling inherent uncertainty, incompleteness and inconsistency in autonomous intelligent systems. The key idea of the solution is to learn knowledge simultaneously at many levels of abstraction using advanced Machine Learning techniques. Our methodology is first to find appropriate knowledge at the optimal level of abstractions from each data source, and then to present the knowledge to a meta-agent equipped with capabilities of reasoning under uncertainty and resolving inconsistency and conflict to autonomously achieve situation awareness and eventually to support decision making in the system.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 07, 2019
- Source ID
- N629091912031
Entities
People
- Nam Huynh
Organizations
- Japan Advanced Institute of Science and Technology
- Office of Naval Research
- United States Navy