Source-linked AI summary
Fast machine-learning online optimization of ultra-cold-atom experiments
P. B. Wigley, P. J. Everitt, A. van den Hengel, J. W. Bastian, M. A. Sooriyabandara, G. D. McDonald, K. S. Hardman, C. D. Quinlivan, P. Manju, C. C. N. Kuhn, I. R. Petersen, A. Luiten, J. J. Hope, N. P. Robins, M. R. Hush
TL;DR
Standard evaporative-cooling theory can oversimplify BEC production dynamics and miss more effective evaporation methods. This paper applies online Gaussian-process optimization, which converged in 10 experiments after 20 shared training runs, compared with 145 runs for Nelder–Mead.
Problem
Microscopic semiclassical theory can oversimplify evaporative cooling and miss more effective methods for producing Bose-Einstein condensates.
Source: p. 1Method
The paper uses online Gaussian-process optimization to model experimental performance and plan future parameter settings.
Source: pp. 1, 3Results
After 20 shared training runs, the machine-learning algorithm converged in 10 experiments, versus 145 runs for Nelder–Mead.
Source: p. 4Takeaways & Limitations
The learned cost landscape can help identify parameters contributing to BEC production and inform experimental design.
Source: p. 5Other parameter correlation lengths were not estimated well, preventing useful cost-landscape predictions and reliable identification of least-important parameters.
Source: p. 4
Abstract
from arXiv · showhide
Machine-designed control of complex devices or experiments can discover strategies superior to those developed via simplified models. We describe an online optimization algorithm based on Gaussian processes and apply it to optimization of the production of Bose-Einstein condensates (BEC). BEC is typically created with an exponential evaporation ramp that is approximately optimal for s-wave, ergodic dynamics with two-body interactions and no other loss rates, but likely sub-optimal for many real experiments. Machine learning using a Gaussian process, in contrast, develops a statistical model of the relationship between the parameters it controls and the quality of the BEC produced. This is an online process, and an active one, as the Gaussian process model updates on the basis of each subsequent experiment and proposes a new set of parameters as a result. We demonstrate that the Gaussian process machine learner is able to discover a ramp that produces high quality BECs in 10 times fewer iterations than a previously used online optimization technique. Furthermore, we show the internal model developed can be used to determine which parameters are essential in BEC creation and which are unimportant, providing insight into the optimization process.
APPENDIX
The appendix outlines Gaussian-process evaluation and likelihood-based hyperparameter fitting, then describes the simple and complex evaporation-ramp parameterizations used in optimization.
Gaussian process evaluation
- The Gaussian process is fit to experimental parameter sets, measured costs, and uncertainties using a constant offset and squared-exponential covariance, with hyperparameters chosen by maximizing the likelihood of the costs.The likelihood is defined as P(C|X, U, H).Source: APPENDIX · p. 6
Parameterizations of evaporation ramps
- Simple ramps specify start and end amplitudes and duration, while complex ramps extend this form with three polynomial terms; shape parameters are independent across ramps, but final time is shared.The added terms correspond to the 3rd-, 4th-, and 5th-order polynomials.Source: APPENDIX · p. 6