Q Memory based active learning for optimizing noisy continuous functions

This paper introduces a new algorithm Q for optimizing the expected output of a multi input noisy continuous function Q is de signed to need only a few experiments it avoids strong assumptions on the form of the function and it is autonomous in that it re quires little problem speci c tweaking These capabilities are directly applicable to industrial processes and may become in creasingly valuable elsewhere as the machine learning eld expands beyond prediction and function identi cation and into embedded active learning subsystems in robots vehicles and consumer products Four existing approaches to this problem re sponse surface methods numerical optimiza tion supervised learning and evolutionary methods all have inadequacies when the re quirement of black box behavior is com bined with the need for few experiments Q uses instance based determination of a con vex region of interest for performing exper iments In conventional instance based ap proaches to learning a neighborhood was de ned by proximity to a query point In con trast Q de nes the neighborhood by a new geometric procedure that captures the size and shape of the zone of possible optimum locations Q also optimizes weighted com binations of outputs and nds inputs to pro duce target outputs We compare Q with other optimizers of noisy functions on several problems includ ing a simulated noisy process with both non linear continuous dynamics and discrete event queueing components Results are en couraging in terms of both speed and auton omy ACTIVE LEARNING FOR