On the Power of Learning-Augmented BSTs

We present the first Learning-Augmented Binary Search Tree(BST) that attains Static Optimality and Working-Set Bound given rough predictions. Following the recent studies in algorithms with predictions and learned index structures, Lin, Luo, and Woodruff (ICML 2022) introduced the concept of Learning-Augmented BSTs, which aim to improve BSTs with learned advice. Unfortunately, their construction gives only static optimality under strong assumptions on the input. In this paper, we present a simple BST maintenance scheme that benefits from learned advice. With proper predictions, the scheme achieves Static Optimality and Working-Set Bound, respectively, which are important performance measures for BSTs. Moreover, the scheme is robust to prediction errors and makes no assumption on the input.