DynaSemble
Dynamic ensembling for better model performance and robustness
Abstract
Ensemble methods combine multiple models to achieve better performance than individual models. However, traditional ensembles use static combinations that don’t adapt to different inputs. We introduce DynaSemble, a dynamic ensembling approach that intelligently selects and weights ensemble members based on input characteristics. Our method learns to predict which models will perform best for specific inputs, leading to improved accuracy and robustness across diverse datasets.