AutoOED: Automated Optimal Experimental Design Platform with Data- and Time-Efficient Multi-Objective Optimization

Yunsheng Tian, Mina Konakovic Lukovic, Michael Foshey, Timothy Erps, Beichen Li, Wojciech Matusik

Keywords: optimal experiment design, bayesian optimization, multi-objective optimization, software platform

Abstract: We present AutoOED, an Automated Optimal Experimental Design platform powered by machine learning to accelerate discovering solutions with optimal objective trade-offs. To solve expensive multi-objective problems in a data-efficient manner, we implement popular multi-objective Bayesian optimization (MOBO) algorithms with state-of-the-art performance in a modular framework. To further accelerate the optimization in a time-efficient manner, we propose a novel strategy called Believer-Penalizer (BP), which allows batch experiments to be accelerated asynchronously without affecting performance. AutoOED serves as a testbed for machine learning researchers to quickly develop and evaluate their own MOBO algorithms. We also provide a graphical user interface (GUI) for users with little or no experience with coding, machine learning, or optimization to visualize and guide the experiment design intuitively. Finally, we demonstrate that AutoOED can control and guide real-world hardware experiments in a fully automated way without human intervention.

One-sentence Summary: AutoOED: an Automated Optimal Experimental Design platform powered by machine learning to accelerate discovering solutions with optimal objective trade-offs.