We propose an information-theoretic initialization strategy for Bayesian optimization that jointly improves predictive accuracy and hyperparameter learning, enabling better surrogate models and optimization efficiency in few-shot settings..