Sample Complexity and Decision-Theoretic Guarantees for Bayesian Model Averaging over Decision Trees with Catalan-Exponential Priors
Researchers have published a paper detailing sample complexity and decision-theoretic guarantees for Bayesian Model Averaging (BMA) over decision trees. The work focuses on determining when BMA weights provide enough information to justify exploiting the averaging distribution. The study provides a complete non-asymptotic theory for rational commitment thresholds, specifically for Bayesian decision trees with Dirichlet-Multinomial leaf models and a Catalan-exponential tree-size prior. AI
IMPACT Provides theoretical underpinnings for decision tree models, potentially improving their sample efficiency and decision-making capabilities.