Inverse Bayesian Optimization: Learning Human Acquisition Functions in an Exploration vs Exploitation Search Task

This paper introduces a probabilistic framework to estimate parame-ters of an acquisition function given observed human behavior that can be mod-eled as a collection of sample paths from a Bayesian optimization procedure. The methodology involves defining a likelihood on observed human behavior from an optimization task, where the likelihood is parameterized by a Bayesian optimiza-tion subroutine governed by an unknown acquisition function. This structure en-ables us to make inference on a subject's acquisition function while allowing their behavior to deviate around the solution to the Bayesian optimization subroutine. To test our methods, we designed a sequential optimization task which forced subjects to balance exploration and exploitation in search of an invisible target location. Applying our proposed methods to the resulting data, we find that many subjects tend to exhibit exploration preferences beyond that of standard acqui-sition functions to capture. Guided by the model discrepancies, we augment the candidate acquisition functions to yield a superior fit to the human behavior in this task.

Automated summary

This paper presents a probabilistic framework for estimating parameters of an acquisition function based on observed human behavior. By defining a likelihood function on the observed behavior, which is parameterized by a Bayesian optimization subroutine, it enables inference on an individual's acquisition function while accounting for deviations in behavior around the optimization solution.