OptiCore: Scalable Greedy Coreset Optimization Method for Efficient Deep Learning

The growing use of Deep Learning in various domains has amplifiedthe challenges of training models due to high computational costsand the need for large volumes of data. To address these limitations,this study presents the OptiCore method, a new dataset optimizationapproach based on the Greedy Coreset technique. OptiCorestrategically reduces the size of datasets while preserving theirrepresentativeness and diversity, integrating computational costanalyses through the Relative Cost Normalized metric. This methodbalances data efficiency and model performance, offering a scalablesolution for practical applications. The methodology is designedfor generalization and reproducibility, extending its usefulness todifferent Deep Learnig contexts. In the case study, Deep Learningmodels were applied for the classification of three-dimensionalshapes, with the ResNet-50 architecture showing the best results.OptiCore reduced the dataset by up to 90%, maintaining competitiveaccuracy while significantly reducing computational demands.