Distributional Learning

Distributional learning focuses on representing and learning from data as probability distributions, rather than single point estimates, to capture uncertainty and improve model robustness. Current research emphasizes developing algorithms and model architectures, such as distributional random forests, that effectively handle multivariate data and incorporate distributional information into various machine learning tasks, including regression, classification, and reinforcement learning. This approach leads to more accurate predictions, better uncertainty quantification, and improved generalization performance across diverse applications, ranging from electricity price forecasting to robotics and graph neural networks.