Ensemble Prediction
Ensemble prediction methods combine multiple models to improve accuracy, robustness, and uncertainty quantification in diverse prediction tasks. Current research emphasizes dynamic weighting of model predictions, exploring architectures like neural networks and diffusion probabilistic models, and employing techniques like sharpness minimization and weighted scoring rules to enhance performance, particularly for extreme events and noisy data. These advancements are significant for improving the reliability and interpretability of predictions across various fields, from weather forecasting and materials science to medical diagnostics and natural language processing.
Papers
Improving probabilistic forecasts of extreme wind speeds by training statistical post-processing models with weighted scoring rules
Jakob Benjamin Wessel, Christopher A. T. Ferro, Gavin R. Evans, Frank Kwasniok
U-learning for Prediction Inference via Combinatory Multi-Subsampling: With Applications to LASSO and Neural Networks
Zhe Fei, Yi Li