Explainable contextual data driven fusion
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
Numerous applications require the intelligent combining of disparate sensor data streams to create a more complete and enhanced observation in support of underlying tasks like classification, regression, or decision making. This presentation is focused on two underappreciated and often overlooked parts of information fusion, explainability and context. Due to the rapidly increasing deployment and complexity of machine learning solutions, it is critical that the humans who deploy these algorithms can understand why and how a given algorithm works, as well as be able to determine when an algorithm is suitable for use in a particular instance of the problem. The first half of this paper outlines a new similarity measure for capacities and integrals. This measure is used to compare machine learned fusion solutions and explain what a single fusion solution learned. The second half of the paper is focused on contextual fusion with respect to incomplete (limited knowledge) models and metadata for unmanned aerial vehicles (UAVs). Example UAV metadata includes platform (e.g., GPS, IMU, etc.) and environmental (e.g., weather, solar position, etc.) data. Incomplete models herein are a result of limitations of machine learning related to under-sampling of training data. To address these challenges, a new contextually adaptive online Choquet integral is outlined.
Table of Contents
DOI
PubMed ID
Degree
M.S.