Guiding exploration of the environment via disagreement is a common approach in model-based ensemble learning methods. For the effectiveness of the approach, the ensemble diversity is a crucial property. In this work, we investigate the usage of a state-of-the-art multi-modal optimizer to generate a diverse ensemble of locally optimal model parameter solutions. We expect to generate more constructive disagreements between the models compared to commonly used methods like random model parameter initialization and bootstrapping. In simulation-to-simulation experiments, we aim to compare diversity strategies based on measures of model diversity and the progress on minimizing the reality gap.
Vortragsdetails
Exploration via Constructive Disagreement: Diversity in Ensemble-based Behavior Learning using Niching Migratory Multi-Swarm Optimization
In der Regel sind die Vorträge Teil von Lehrveranstaltungsreihen der Universität Bremen und nicht frei zugänglich. Bei Interesse wird um Rücksprache mit dem Sekretariat unter sek-ric(at)dfki.de gebeten.
