Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:1805.00654 · Nature Communications

Multiparameter optimisation of a magneto-optical trap using deep learning

Aaron D. Tranter, Harry J. Slatyer, Michael Hush, Anthony C. Leung, Jesse L. Everett, Karun Paul, Pierre Vernaz-Gris, Ping Koy Lam, B. C. Buchler, Geoff Campbell

Published Oct 15, 2018• 95 citations
arXivHTMLPDF

Abstract

Machine learning based on artificial neural networks has emerged as an efficient means to develop empirical models of complex systems. Cold atomic ensembles have become commonplace in laboratories around the world, however, many-body interactions give rise to complex dynamics that preclude precise analytic optimisation of the cooling and trapping process. Here, we implement a deep artificial neural network to optimise the magneto-optic cooling and trapping of neutral atomic ensembles. The solution identified by machine learning is radically different to the smoothly varying adiabatic solutions currently used. Despite this, the solutions outperform best known solutions producing higher optical densities.

DOI
10.1038/s41467-018-06847-1

Related papers

Multiparameter optimisation of a magneto-optical trap using deep learning

Aaron D. Tranter, Harry J. Slatyer, Michael Hush +7 more

Abstract Machine learning based on artificial neural networks has emerged as an efficient means to develop empirical models of complex systems. Cold atomic ensembles have become commonplace in laboratories around the world, however, many-body interactions give rise to complex dynamics that preclude precise analytic optimisation of the cooling and trapping process. Here, we implement a deep artificial neural network to optimise the magneto-optic cooling and trapping of neutral atomic ensembles. The solution identified by machine learning is radically different to the smoothly varying adiabatic solutions currently used. Despite this, the solutions outperform best known solutions producing higher optical densities.

•Jan 1, 2018•75 citations

Machine learning for quantum and complex systems

Aaron D. Tranter

Machine learning now plays a pivotal role in our society, providing solutions to problems that were previously thought intractable. The meteoric rise of this technology can no doubt be attributed to the information age that we now live in. As data is continually amassed, more efficient and scalable methods are required to yield functional models and accurate inferences. Simultaneously we have also seen quantum technology come to the forefront of research and next generation systems. These technologies promise secure information transfer, efficient computation and high precision sensing, at levels unattainable by their classical counterparts. Although these technologies are powerful, they are necessarily more complicated and difficult to control. The combination of these two advances yields an opportunity for study, namely leveraging the power of machine learning to control and optimise quantum (and more generally complex) systems. The work presented in thesis explores these avenues of investigation and demonstrates the potential success of machine learning methods in the domain of quantum and complex systems. One of the most crucial potential quantum technologies is the quantum memory. If we are to one day harness the true power of quantum key distribution for secure transimission of information, and more general quantum computating tasks, it will almost certainly involve the use of quantum memorys. We start by presenting the operation of the cold atom workhorse: the magneto-optical trap (MOT). To use a cold atomic ensemble as a quantum memory we are required to prepare the atoms using a specialised cooling sequence. During this we observe a stable, coherent optical emission exiting each end of the elongated ensemble. We characterise this behaviour and compare it to similar observations in previous work. Following this, we use the ensemble to implement a backward Raman memory. Using this scheme we are able to demonstrate an increased efficiency over that of previous forward recall implementations. While we are limited by the optical depth of the system, we observe an efficiency more than double that of previous implementations. The MOT provides an easily accessible test bed for the optimisation via some machine learning technique. As we require an efficient search method, we implement a new type of algorithm based on deep learning. We design this technique such that the artificial neural networks are placed in control of the online optimisation, rather than simply being used as surrogate models. We experimentally optimise the optical depth of the MOT using this method, by parametrising the time varying compression sequence. We identify a new and unintuitive method for cooling the atomic ensemble which surpasses current methods. Following this initial implementation we make substantial improvements to the deep learning approach. This extends the approach to be applicable to a far wider range of complex problems, which may contain extensive local minima and structure. We benchmark this algorithm against many of the conventional optimisation techniques and demonstrate superior capability to optimise problems with high dimensionality. Finally we apply this technique to a series of preliminary problems, namely the tuning of a single electron transistor and second-order correlations from a quantum dot source.

•Jan 1, 2021•2 citations

Nondestructive characterization of laser-cooled atoms using machine learning

Guilherme de Sousa, M. Doris, Dario D'Amato +3 more

Laser cooling and trapping techniques are foundational to modern atomic physics, with applications including atomic clocks and sensors as well as quantum simulators and computers. These applications require stable, well-calibrated laser cooling and accurate determination of atom number and temperature. We develop machine learning techniques for estimating physical properties of laser-cooled potassium-39 atoms in a magneto-optical trap using only the scattered light—i.e., fluorescence—that is intrinsic to the cooling process. In situ snapshot images of fluorescing atomic ensembles directly reveal the spatial structure of these millimeter-scale objects but contain no obvious information regarding internal properties such as the temperature. We first assembled and labeled a balanced dataset sampling 8 × 10 3 different experimental parameters that includes examples with large and dense atomic ensembles, a complete absence of atoms, and everything in between. We describe a range of models trained to predict atom number and temperature solely from fluorescence images. These run the gamut from a poorly performing linear regression model based only on integrated fluorescence to deep neural networks that give number and temperature with fractional uncertainties of 0.1 and 0.2, respectively. These results show that fluorescence images can provide fast, nondestructive diagnostics for cold-atom experiments and may enable real-time experimental feedback.

Newton•May 14, 2026•0 citations

Machine learner optimization of optical nanofiber-based dipole traps

Ratnesh Kumar Gupta, Jesse L. Everett, Aaron D. Tranter +4 more

We use a machine learning optimizer to increase the number of rubidium-87 atoms trapped in an optical nanofiber-based two-color evanescent dipole trap array. Collisional blockade limits the average number of atoms per trap to about 0.5, and a typical uncompensated rubidium trap has even lower occupancy due to challenges in simultaneously cooling atoms and loading them in the traps. Here, we report on the implementation of an in-loop stochastic artificial neural network machine learner to optimize this loading by optimizing the absorption of a near-resonant, nanofiber-guided, probe beam. By giving the neural network control of the laser cooling process, we observe an increase in peak optical depth of 66% from 3.2 ± 0.2 to 5.3 ± 0.3. We use a microscopic model of the atomic absorption to infer an increase in the number of dipole-trapped atoms from 300 ± 60 to 450 ± 90 and a small decrease in their average temperature from 150 to 140 μK. The machine learner is able to quickly and effectively explore the large parameter space of the laser cooling control process so as to find optimal parameters for loading the dipole traps. The increased number of atoms should facilitate studies of collective atom–light interactions mediated via the evanescent field.

arXiv:2110.03931•AVS Quantum Science•May 11, 2022•22 citations

Reinforcement learning in cold atom experiments

Malte Reinschmidt, József Fortágh, Andreas Günther +1 more

Cold atom traps are at the heart of many quantum applications in science and technology. The preparation and control of atomic clouds involves complex optimization processes, that could be supported and accelerated by machine learning. In this work, we introduce reinforcement learning to cold atom experiments and demonstrate a flexible and adaptive approach to control a magneto-optical trap. Instead of following a set of predetermined rules to accomplish a specific task, the objectives are defined by a reward function. This approach not only optimizes the cooling of atoms just as an experimentalist would do, but also enables new operational modes such as the preparation of pre-defined numbers of atoms in a cloud. The machine control is trained to be robust against external perturbations and able to react to situations not seen during the training. Finally, we show that the time consuming training can be performed in-silico using a generic simulation and demonstrate successful transfer to the real world experiment.

Nature Communications•Oct 2, 2024•13 citations

Reinforcement Learning in Ultracold Atom Experiments

Malte Reinschmidt, József Fortágh, Andreas Günther +1 more

Cold atom traps are at the heart of many quantum applications in science and technology. The preparation and control of atomic clouds involves complex optimization processes, that could be supported and accelerated by machine learning. In this work, we introduce reinforcement learning to cold atom experiments and demonstrate a flexible and adaptive approach to control a magneto-optical trap. Instead of following a set of predetermined rules to accomplish a specific task, the objectives are defined by a reward function. This approach not only optimizes the cooling of atoms just as an experimentalist would do, but also enables new operational modes such as the preparation of pre-defined numbers of atoms in a cloud. The machine control is trained to be robust against external perturbations and able to react to situations not seen during the training. Finally, we show that the time consuming training can be performed in-silico using a generic simulation and demonstrate successful transfer to the real world experiment.

arXiv:2306.16764•Jun 29, 2023•3 citations
Machine learning based on artificial neural networks has emerged as an efficient means to develop empirical models of complex systems. Cold atomic ensembles have become commonplace in laboratories around the world, however, many-body interactions give rise to complex dynamics that preclude precise analytic optimisation of the cooling and trapping process. Here, we implement a deep artificial neural network to optimise the magneto-optic cooling and trapping of neutral atomic ensembles. The solution identified by machine learning is radically different to the smoothly varying adiabatic solutions currently used. Despite this, the solutions outperform best known solutions producing higher optical densities.
Abstract Machine learning based on artificial neural networks has emerged as an efficient means to develop empirical models of complex systems. Cold atomic ensembles have become commonplace in laboratories around the world, however, many-body interactions give rise to complex dynamics that preclude precise analytic optimisation of the cooling and trapping process. Here, we implement a deep artificial neural network to optimise the magneto-optic cooling and trapping of neutral atomic ensembles. The solution identified by machine learning is radically different to the smoothly varying adiabatic solutions currently used. Despite this, the solutions outperform best known solutions producing higher optical densities.
Machine learning now plays a pivotal role in our society, providing solutions to problems that were previously thought intractable. The meteoric rise of this technology can no doubt be attributed to the information age that we now live in. As data is continually amassed, more efficient and scalable methods are required to yield functional models and accurate inferences. Simultaneously we have also seen quantum technology come to the forefront of research and next generation systems. These technologies promise secure information transfer, efficient computation and high precision sensing, at levels unattainable by their classical counterparts. Although these technologies are powerful, they are necessarily more complicated and difficult to control. The combination of these two advances yields an opportunity for study, namely leveraging the power of machine learning to control and optimise quantum (and more generally complex) systems. The work presented in thesis explores these avenues of investigation and demonstrates the potential success of machine learning methods in the domain of quantum and complex systems. One of the most crucial potential quantum technologies is the quantum memory. If we are to one day harness the true power of quantum key distribution for secure transimission of information, and more general quantum computating tasks, it will almost certainly involve the use of quantum memorys. We start by presenting the operation of the cold atom workhorse: the magneto-optical trap (MOT). To use a cold atomic ensemble as a quantum memory we are required to prepare the atoms using a specialised cooling sequence. During this we observe a stable, coherent optical emission exiting each end of the elongated ensemble. We characterise this behaviour and compare it to similar observations in previous work. Following this, we use the ensemble to implement a backward Raman memory. Using this scheme we are able to demonstrate an increased efficiency over that of previous forward recall implementations. While we are limited by the optical depth of the system, we observe an efficiency more than double that of previous implementations. The MOT provides an easily accessible test bed for the optimisation via some machine learning technique. As we require an efficient search method, we implement a new type of algorithm based on deep learning. We design this technique such that the artificial neural networks are placed in control of the online optimisation, rather than simply being used as surrogate models. We experimentally optimise the optical depth of the MOT using this method, by parametrising the time varying compression sequence. We identify a new and unintuitive method for cooling the atomic ensemble which surpasses current methods. Following this initial implementation we make substantial improvements to the deep learning approach. This extends the approach to be applicable to a far wider range of complex problems, which may contain extensive local minima and structure. We benchmark this algorithm against many of the conventional optimisation techniques and demonstrate superior capability to optimise problems with high dimensionality. Finally we apply this technique to a series of preliminary problems, namely the tuning of a single electron transistor and second-order correlations from a quantum dot source.
Laser cooling and trapping techniques are foundational to modern atomic physics, with applications including atomic clocks and sensors as well as quantum simulators and computers. These applications require stable, well-calibrated laser cooling and accurate determination of atom number and temperature. We develop machine learning techniques for estimating physical properties of laser-cooled potassium-39 atoms in a magneto-optical trap using only the scattered light—i.e., fluorescence—that is intrinsic to the cooling process. In situ snapshot images of fluorescing atomic ensembles directly reveal the spatial structure of these millimeter-scale objects but contain no obvious information regarding internal properties such as the temperature. We first assembled and labeled a balanced dataset sampling 8 × 10 3 different experimental parameters that includes examples with large and dense atomic ensembles, a complete absence of atoms, and everything in between. We describe a range of models trained to predict atom number and temperature solely from fluorescence images. These run the gamut from a poorly performing linear regression model based only on integrated fluorescence to deep neural networks that give number and temperature with fractional uncertainties of 0.1 and 0.2, respectively. These results show that fluorescence images can provide fast, nondestructive diagnostics for cold-atom experiments and may enable real-time experimental feedback.
We use a machine learning optimizer to increase the number of rubidium-87 atoms trapped in an optical nanofiber-based two-color evanescent dipole trap array. Collisional blockade limits the average number of atoms per trap to about 0.5, and a typical uncompensated rubidium trap has even lower occupancy due to challenges in simultaneously cooling atoms and loading them in the traps. Here, we report on the implementation of an in-loop stochastic artificial neural network machine learner to optimize this loading by optimizing the absorption of a near-resonant, nanofiber-guided, probe beam. By giving the neural network control of the laser cooling process, we observe an increase in peak optical depth of 66% from 3.2 ± 0.2 to 5.3 ± 0.3. We use a microscopic model of the atomic absorption to infer an increase in the number of dipole-trapped atoms from 300 ± 60 to 450 ± 90 and a small decrease in their average temperature from 150 to 140 μK. The machine learner is able to quickly and effectively explore the large parameter space of the laser cooling control process so as to find optimal parameters for loading the dipole traps. The increased number of atoms should facilitate studies of collective atom–light interactions mediated via the evanescent field.
Cold atom traps are at the heart of many quantum applications in science and technology. The preparation and control of atomic clouds involves complex optimization processes, that could be supported and accelerated by machine learning. In this work, we introduce reinforcement learning to cold atom experiments and demonstrate a flexible and adaptive approach to control a magneto-optical trap. Instead of following a set of predetermined rules to accomplish a specific task, the objectives are defined by a reward function. This approach not only optimizes the cooling of atoms just as an experimentalist would do, but also enables new operational modes such as the preparation of pre-defined numbers of atoms in a cloud. The machine control is trained to be robust against external perturbations and able to react to situations not seen during the training. Finally, we show that the time consuming training can be performed in-silico using a generic simulation and demonstrate successful transfer to the real world experiment.
Cold atom traps are at the heart of many quantum applications in science and technology. The preparation and control of atomic clouds involves complex optimization processes, that could be supported and accelerated by machine learning. In this work, we introduce reinforcement learning to cold atom experiments and demonstrate a flexible and adaptive approach to control a magneto-optical trap. Instead of following a set of predetermined rules to accomplish a specific task, the objectives are defined by a reward function. This approach not only optimizes the cooling of atoms just as an experimentalist would do, but also enables new operational modes such as the preparation of pre-defined numbers of atoms in a cloud. The machine control is trained to be robust against external perturbations and able to react to situations not seen during the training. Finally, we show that the time consuming training can be performed in-silico using a generic simulation and demonstrate successful transfer to the real world experiment.