How to Evaluate High Level Fusion Algorithms?

Evaluating high level information fusion algorithms is a tricky problem. Most of the time, situations monitored through high level information fusion are complex, composed of multiple objects or entities, having heterogeneous properties and in relation with each other's. Many criteria have to b...

Full description

Saved in:
Bibliographic Details
Published in2019 22th International Conference on Information Fusion (FUSION) pp. 1 - 8
Main Authors Laudy, Claire, Museux, Nicolas
Format Conference Proceeding
LanguageEnglish
Published ISIF - International Society of Information Fusion 01.07.2019
Subjects
Aggregates
artificial intelligence
Benchmark testing
Encyclopedias
evaluation
high level fusion
Machine learning algorithms
Measurement
metrics
Ontologies
Uncertainty
Online AccessGet full text
DOI10.23919/FUSION43075.2019.9011206

Cover

More Information
Summary:Evaluating high level information fusion algorithms is a tricky problem. Most of the time, situations monitored through high level information fusion are complex, composed of multiple objects or entities, having heterogeneous properties and in relation with each other's. Many criteria have to be taken into account within the evaluation. In this paper, we define several performance evaluation criteria. We focus on criteria related to functional evaluation, namely the correctness, the completeness and the precision of the result, as well as the level of management of uncertainty of information. Our criteria rely on the comparison of the result, given by the evaluated fusion algorithm, with the expected result of a given set of information provided as an input benchmark. We then present a proposition to aggregate them together with the 2-additive Choquet integral to obtain a single evaluation score.
DOI:10.23919/FUSION43075.2019.9011206