Single-Exposure Absorption Imaging of Ultracold Atoms Using Deep Learning

Ness, Gal; Vainbaum, Anastasiya; Shkedrov, Constantine; Florshaim, Yanay; Sagi, Yoav

doi:10.1103/physrevapplied.14.014011

Cited by 28 publications

(24 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other ML techniques have the potential to further improve our result. In Ness et al [20], deep learning is used to predict the background noise based on the signal surrounding the atoms. A similar approach may allow our procedure to forgo subtracting the second background image.…”

Section: Discussionmentioning

confidence: 99%

Utilizing machine learning to improve the precision of fluorescence imaging of cavity-generated spin squeezed states

Malia,

Wu,

Martínez-Rincón

et al. 2021

Preprint

View full text Add to dashboard Cite

We present a supervised learning model to calibrate the photon collection rate during the fluorescence imaging of cold atoms. The linear regression model finds the collection rate at each location on the sensor such that the atomic population difference equals that of a highly precise optical cavity measurement. This 192 variable regression results in a measurement variance 27% smaller than our previous single variable regression calibration. The measurement variance is now in agreement with the theoretical limit due to other known noise sources. This model efficiently trains in less than a minute on a standard personal computer's CPU, and requires less than 10 minutes of data collection. Furthermore, the model is applicable across a large changes in population difference and across data collected on different days.

show abstract

Section: Discussionmentioning

confidence: 99%

Utilizing machine learning to improve the precision of fluorescence imaging of cavity-generated spin squeezed states

Malia,

Wu,

Martínez-Rincón

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…As a result, the absorption signal is weak. To improve the signal-to-noise ratio, we employ a deep-learning approach to filter out the background noise in the images [106]. We have verified that this noise removal procedure does not change significantly the reported CF values and only reduces the uncertainty.…”

Section: Appendix B: Extracting the Condensate Fractionmentioning

confidence: 93%

Absence of heating in a uniform Fermi gas created by periodic driving

Shkedrov,

Menashes,

Ness

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Ultracold atoms are a powerful resource for quantum technologies. As such, they are usually confined in an external potential that often depends on the atomic spin, which may lead to inhomogeneous broadening, phase separation and decoherence. Dynamical decoupling provides an approach to mitigate these effects by applying an external field that induces rapid spin rotations. However, a continuous periodic driving of a generic interacting many-body system eventually heats it up. The question is whether dynamical decoupling can be applied at intermediate times without altering the underlying physics. Here we answer this question affirmatively for a strongly interacting degenerate Fermi gas held in a flat box-like potential. We counteract most of the gravitational force by applying an external magnetic field with an appropriate gradient. Since the magnetic force, and consequently, the whole potential, is spin-dependent, we employ rf to induce a rapid spin rotation. The driving causes atoms in both spin states to experience the same time-average flat potential, leading to a uniform cloud. Most importantly, we find that when the driving frequency is high enough, there is no heating on experimentally relevant timescales, and physical observables are similar to those of a stationary gas. In particular, we measure the pair-condensation fraction of a fermionic superfluid at unitarity and the contact parameter in the BEC-BCS crossover. The condensate fraction exhibits a non-monotonic dependence on the drive frequency and reaches a value higher than its value without driving. The contact agrees with recent theories and calculations for a uniform stationary gas. Our results establish that a strongly-interacting quantum gas can be dynamically decoupled from a spin-dependent potential for long periods of time without modifying its intrinsic many-body behavior.

show abstract

“…From these images a few physical quantities can be extracted directly, whereas others require fitting with effective theories. One of neural networks greatest strengths is image processing and pattern recognition, which can be leveraged to directly extract information from the images [24] without having to fit the data with appropriate models. Hence, neural networks can potentially yield an efficiency gain, as more information can be extracted per image using neural networks compared to fitting.…”

Section: Introductionmentioning

confidence: 99%

Thermometry of one-dimensional Bose gases with neural networks

Møller,

Schweigler,

Tajik

et al. 2021

Preprint

View full text Add to dashboard Cite

Thermometry of one-dimensional Bose gases can be achieved in various ways, which all revolve around comparing measurement statistics to theoretical prediction. As a result, many repetitions of the measurement are required before an accurate comparison can be made. In this work, we train a neural network to estimate the temperature of a one-dimensional Bose gas in the quasi-condensate regime from a single absorption image. We benchmark our model on both simulated and experimentally measured data. Comparing with our standard method of density ripple thermometry based on fitting two-point correlation function shows that the network can achieve the same precision needing only half the amount of images. Our findings highlight the gain in efficiency when incorporating neural networks into analysis of data from cold gas experiments. Further, the neural network architecture presented can easily be reconfigured to extract other parameters from the images.

show abstract

Single-Exposure Absorption Imaging of Ultracold Atoms Using Deep Learning

Cited by 28 publications

References 14 publications

Utilizing machine learning to improve the precision of fluorescence imaging of cavity-generated spin squeezed states

Utilizing machine learning to improve the precision of fluorescence imaging of cavity-generated spin squeezed states

Absence of heating in a uniform Fermi gas created by periodic driving

Thermometry of one-dimensional Bose gases with neural networks

Contact Info

Product

Resources

About