Trust Estimation in Forecasting-Based Knowledge Fusion

0549011 - ÚTIA 2022 RIV LU eng C - Conference Paper (international conference)
Kárný, Miroslav - Karlík, Daniel
Trust Estimation in Forecasting-Based Knowledge Fusion.
Proceedings of BNAIC/BeneLearn 2021. Luxembourg: University of Luxembourg, 2021 - (Leiva, L.; Pruski, C.; Markovich, R.; Najjar, A.; Schommer, C.), s. 363-378
[Benelux Conference on Artificial Intelligence and 30th Belgian-Dutch Conference on Machine Learning 2021 /33./. Belval, Esch-sur-Alzette (LU), 10.11.2021-12.11.2021]
R&D Projects: GA MŠMT(CZ) LTC18075
Grant - others:COST (European Cooperation in Science and Technology)(XE) CA 16228
Institutional support: RVO:67985556
Keywords : Trust * Knowledge sharing * Forecasting * Fusion * Decision making * Bayesianism
OECD category: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
http://library.utia.cas.cz/separaty/2021/AS/karny-0549011.pdf

Inference and decision making (DM) are ultimate goals of the artificialintelligence use. Complexity of DM tasks is the main barrier of their efficient solutions. Complex tasks are solved by dividing them among cooperating agents. This requires a knowledge fusion at a solution stage. It always has to cope with uncertainty. The used Bayesianism quantifies the uncertain knowledge by a probability density (pd) of modelled variables. The knowledge accumulation evolves the posterior pd of a parameter in the parametric model of observations. Bayes’rule updates the posterior pd. It provides a lossless compression of the knowledge in the observed data. An extended Bayes’ rule enables the use of knowledge coded in a forecaster of the modelled observations supplied by an agent’sneighbour. This rule exploits a weight expressing the trust into the forecaster. The paper offers yet-missing, algorithmic, data-based choice of this weight. It applies Bayesian estimation while assuming an invariant trust weight. Simulated examples illustrate behaviour of the resulting algorithm. They inspect its sensitivity to violation of the assumed credibility invariance. This prepares solutions coping with volatile knowledge sources.
Permanent Link: http://hdl.handle.net/11104/0325137