Efficient Estimation of Hessian Matrices for Stochastic Learning to Rank with Gradient Boosted Trees

Introducing a novel estimator for the second-order derivatives (Hessian matrix) of stochastic ranking objectives, enabling effective optimization of Gradient Boosted Decision Trees (GBDTs) for learning to rank tasks.