Suche
Anzeige der Treffer 1-4 von 4
Bereiche und Sammlungen, die Ihrer Suche entsprechen.
This research data serves as verification of our experiments and specifically includes the checkpoints during the training of our models via (multi-agent) reinforcement learning, allowing us to reconstruct the progress of the training. The weights of the trained models were serialized using the Python module pickle, and can also be reloaded using this module. In this way, the state of the model can be restored at fixed time steps (checkpoints), for example to continue the training, to evaluate the model and/or to document/archive the progress.
Elastic displays afford a natural stacking of information layers in their haptic interaction space. So far, research has focused on technical issues and interaction concepts by pushing and pulling the surface of such displays. We report a study with 24 participants that investigates the feasibility and limitations of interaction with 6 to 21 layers. We measured completion times and error rates for reaching and holding specific target layers and observed performed gestures. Our findings show that solution time increases with more layers while precision is maintained with a mean success rate of holding a layer of 70 %, even when using up to 21 layers. User experience measurements show that hedonic qualities of the interaction were rated higher than pragmatic qualities. Our investigation proves the general feasibility of using layer-based interaction with an elastic display and provides guidelines on the limitations.
The evaluation datasets stored in this collection are used to compare the performance of multi-agent reinforcement learning. In addition, the performance of other methods such as (meta-) heuristic algorithms or single agent reinforcement learning algorithms or novel methods of search space reduction can also be compared.