Bayesian Optimization for Dynamic Pricing and Learning

We study the dynamic pricing problem in which a seller, facing an unknown stochastic demand function, must adaptively choose prices to maximize cumulative revenue. The seller observes only noisy demand observations after posting prices, leading to the need for simultaneous learning of the demand model and the optimization of pricing decisions. Existing methods typically rely on parametric assumptions (e.g., polynomial or logit demand) and optimize using Iterative Least Squares or reinforcement learning (RL) algorithms. These approaches are vulnerable to model misspecification, slow convergence, and high variance under noise.