Applying machine learning optimization methods to the production of a quantum gas

A J Barker,K Luksch,C J Foot,H Style,S Sunami,F Hill,D Garrick,E Bentine

doi:10.1088/2632-2153/ab6432

Abstract

We apply three machine learning strategies to optimize the atomic cooling processes utilized in the production of a Bose–Einstein condensate (BEC). For the first time, we optimize both laser cooling and evaporative cooling mechanisms simultaneously. We present the results of an evolutionary optimization method (differential evolution), a method based on non-parametric inference (Gaussian process regression) and a gradient-based function approximator (artificial neural network). Online optimization is performed using no prior knowledge of the apparatus, and the learner succeeds in creating a BEC from completely randomized initial parameters. Optimizing these cooling processes results in a factor of four increase in BEC atom number compared to our manually-optimized parameters. This automated approach can maintain close-to-optimal performance in long-term operation. Furthermore, we show that machine learning techniques can be used to identify the main sources of instability within the apparatus.

Highlights

Recent developments in artificial intelligence and machine learning have provided tools with which a computer can outperform the analytic capability of a human, when data sets are large or when a system relies on many free parameters [1]
Optimization is performed using no prior knowledge of the apparatus, and the learner succeeds in creating a Bose–Einstein condensate (BEC) from completely randomized initial parameters
We describe our experimental apparatus and the several stages of trapping and cooling of an atomic vapor which lead to the production of a BEC [21, 23]

Summary

Introduction

Recent developments in artificial intelligence and machine learning have provided tools with which a computer can outperform the analytic capability of a human, when data sets are large or when a system relies on many free parameters [1]. Machine learning is well suited to the optimization of a complex experimental apparatus [4,5,6]. As compared to a human, a major advantage of many machine learning methods is that the chosen learner has no preconceptions for how the parameters should affect the final result, and is objectively guided purely by the actual data. We apply three different machine learning algorithms to optimize an atomic physics experiment. Bose–Einstein condensation in a dilute atomic vapor was first realized in 1995, resulting in the award of the Nobel Prize in 2001 [8, 9]. Ultracold atomic vapor experiments have been used to investigate a wide range of physical phenomena, including quantum many-body physics [10], quantum-mechanical phase transitions [11, 12] and superfluid turbulence [13]

Methods

Results

Conclusion