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

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