amortized-learning

The paper proposes an amortized framework for learning bandwidth selection in kernel density estimation under logarithmic score, showing consistent performance gains over classical methods in various sampling scenarios.