Non-Myopic-BO-with-Cost-Constraints

Bayesian Optimization (BO) is a powerful tool for optimizing expensive black-box functions. It treats the objective function as a probabilistic model and iteratively updates this model based on observations, typically using Gaussian Processes (GPs). Bayesian optimization can be used in a setting where evaluation is expensive, say tuning the parameters of a neural network to optimize accuracy. Consider the scenario where each evaluation point has a cost associated to it, and there is a budget of the total cost to optimize the underlying function. Say this cost is unknown a-priori and it must be explored as the objective function is evaluated. This is the setting this paper will explore and propose a methodology to model such a scenario by building a GP around both the cost function and objective function.