Simplex-Constrained Sparse Bagging: Transitioning from Uniform Priors to Sparse Posteriors in Ensemble Learning

The article introduces Simplex-Constrained Sparse Bagging (SCSB), a framework for enhancing bootstrap-based bagging ensembles by transitioning from uniform priors to sparse posteriors. SCSB formulates ensemble pruning and calibration as a joint optimization problem over the probability simplex, achieving up to 96% compression and improved inference speed while lowering Expected Calibration Error. This model-agnostic approach addresses the limitations of traditional bagging methods, making it significant for practitioners seeking to optimize ensemble performance and efficiency in AI applications.