Kuijper, ArjanKrämer, MichelKocon, KevinHoebelt, DennisDennisHoebelt2025-04-012025-04-012025https://publica.fraunhofer.de/handle/publica/486022The pipeline introduced in this paper takes in a coreset method, a dataset a model and a hyperparameter algorithm, which selects a coreset to be used in the selection process. The resulting hyperparameter sets are used to train the model. This results in up to 70% in time savings while the performance of the model does deteriorate only marginal, compared to when using the full dataset for the hyperparameter optimization. These results were obtained using two datasets within the topic of computer vision: cifar10, a classification benchmark, and potsdam, a geo semantic segmnetation task. The coreset method ”Herding” generated the best performing subsets relevant to hyperparameter tuning.enBranche: Information TechnologyBranche: BioeconomyResearch Line: Computer graphics (CG)Research Line: Machine learning (ML)LTA: Machine intelligence, algorithms, and data structures (incl. semantics)Geospatial dataArtificial intelligence (AI)Convolutional Neural Networks (CNN)master thesis