Learning-Augmented Ski Rental with Discrete Distributions: A Bayesian Approach

Bosun Kang, Chenglin Fan

Abstract: This paper studies the multi-shop ski rental problem under distributional uncertainty using a Bayesian learning-augmented framework. We propose a causal Bayesian stopping rule that maintains and updates a discrete belief over season length via survival observations. The algorithm cleanly separates prior initialization from online refinement, achieving provable competitive guarantees while remaining modular to predictor fusion and repeated learning across seasons.