Jane Bromley
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Jane Bromley has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
Hugh Dickinson
School of Physical Sciences, The Open University, Walton Hall, Milton Keynes MK7 6AA, UK
2 shared publications- Joshua Wilde
Institut de Ciències del Cosmos (ICCUB), Universitat de Barcelona (IEEC-UB)
2 shared publications - Stephen Serjeant
School of Physical Sciences, The Open University , Milton Keynes MK7 6AA
2 shared publications - Alejandra Melo
Max-Planck-Institut für Astrophysik
1 shared publication Aleksandra Grudskaia
Max-Planck-Institut für Astrophysik , Karl-Schwarzschild Str 1, D-85748 Garching, Germany
1 shared publicationAlexander Fritz
OmegaLambdaTec GmbH , Lichtenbergstraße 8, D-85748 Garching, Germany
1 shared publication
Latest Publications
- Open Access
- 7 June 2023
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 523 (3) , 4413-4430
The impact of human expert visual inspection on the discovery of strong gravitational lenses
We investigate the ability of human ’expert’ classifiers to identify strong gravitational lens candidates in Dark Energy Survey like imaging. We recruited a total of 55 people that completed more than 25% of the project. During the classification task, we present to the participants 1489 images. The sample contains a variety of data including lens simulations, real lenses, non-lens examples, and unlabeled data. We find that experts are extremely good at finding bright, well-resolved Einstein rings, whilst arcs with g-band signal-to-noise less than ∼25 or Einstein radii less than ∼1.2 times the seeing are rarely recovered. Very few non-lenses are scored highly. There is substantial variation in the performance of individual classifiers, but they do not appear to depend on the classifier’s experience, confidence or academic position. These variations can be mitigated with a team of 6 or more independent classifiers. Our results give confidence that humans are a reliable pruning step for lens candidates, providing pure and quantifiably complete samples for follow-up studies.
- 7 June 2023
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 523 (3) , 4413-4430
Open AccessThe impact of human expert visual inspection on the discovery of strong gravitational lenses
We investigate the ability of human ’expert’ classifiers to identify strong gravitational lens candidates in Dark Energy Survey like imaging. We recruited a total of 55 people that completed more than 25% of the project. During the classification task, we present to the participants 1489 images. The sample contains a variety of data including lens simulations, real lenses, non-lens examples, and unlabeled data. We find that experts are extremely good at finding bright, well-resolved Einstein rings, whilst arcs with g-band signal-to-noise less than ∼25 or Einstein radii less than ∼1.2 times the seeing are rarely recovered. Very few non-lenses are scored highly. There is substantial variation in the performance of individual classifiers, but they do not appear to depend on the classifier’s experience, confidence or academic position. These variations can be mitigated with a team of 6 or more independent classifiers. Our results give confidence that humans are a reliable pruning step for lens candidates, providing pure and quantifiably complete samples for follow-up studies.
- 8 February 2022
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 512 (3) , 3464-3479
Detecting gravitational lenses using machine learning: exploring interpretability and sensitivity to rare lensing configurations
Forthcoming large imaging surveys such as Euclid and the Vera Rubin Observatory Legacy Survey of Space and Time are expected to find more than 105 strong gravitational lens systems, including many rare and exotic populations such as compound lenses, but these 105 systems will be interspersed among much larger catalogues of ∼109 galaxies. This volume of data is too much for visual inspection by volunteers alone to be feasible and gravitational lenses will only appear in a small fraction of these data which could cause a large amount of false positives. Machine learning is the obvious alternative but the algorithms’ internal workings are not obviously interpretable, so their selection functions are opaque and it is not clear whether they would select against important rare populations. We design, build, and train several convolutional neural networks (CNNs) to identify strong gravitational lenses using VIS, Y, J, and H bands of simulated data, with F1 scores between 0.83 and 0.91 on 100 000 test set images. We demonstrate for the first time that such CNNs do not select against compound lenses, obtaining recall scores as high as 76 per cent for compound arcs and 52 per cent for double rings. We verify this performance using Hubble Space Telescope and Hyper Suprime-Cam data of all known compound lens systems. Finally, we explore for the first time the interpretability of these CNNs using Deep Dream, Guided Grad-CAM, and by exploring the kernels of the convolutional layers, to illuminate why CNNs succeed in compound lens selection.
- 8 February 2022
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 512 (3) , 3464-3479
Detecting gravitational lenses using machine learning: exploring interpretability and sensitivity to rare lensing configurations
Forthcoming large imaging surveys such as Euclid and the Vera Rubin Observatory Legacy Survey of Space and Time are expected to find more than 105 strong gravitational lens systems, including many rare and exotic populations such as compound lenses, but these 105 systems will be interspersed among much larger catalogues of ∼109 galaxies. This volume of data is too much for visual inspection by volunteers alone to be feasible and gravitational lenses will only appear in a small fraction of these data which could cause a large amount of false positives. Machine learning is the obvious alternative but the algorithms’ internal workings are not obviously interpretable, so their selection functions are opaque and it is not clear whether they would select against important rare populations. We design, build, and train several convolutional neural networks (CNNs) to identify strong gravitational lenses using VIS, Y, J, and H bands of simulated data, with F1 scores between 0.83 and 0.91 on 100 000 test set images. We demonstrate for the first time that such CNNs do not select against compound lenses, obtaining recall scores as high as 76 per cent for compound arcs and 52 per cent for double rings. We verify this performance using Hubble Space Telescope and Hyper Suprime-Cam data of all known compound lens systems. Finally, we explore for the first time the interpretability of these CNNs using Deep Dream, Guided Grad-CAM, and by exploring the kernels of the convolutional layers, to illuminate why CNNs succeed in compound lens selection.
- Open Access
- 30 July 2021
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 507 (1) , 1546-1556
Superresolving Herschel imaging: a proof of concept using Deep Neural Networks
Wide-field submillimetre surveys have driven many major advances in galaxy evolution in the past decade, but without extensive follow-up observations the coarse angular resolution of these surveys limits the science exploitation. This has driven the development of various analytical deconvolution methods. In the last half a decade Generative Adversarial Networks have been used to attempt deconvolutions on optical data. Here, we present an auto-encoder with a novel loss function to overcome this problem in the submillimeter wavelength range. This approach is successfully demonstrated on Herschel SPIRE 500 $\mu\mathrm{m}$ COSMOS data, with the superresolving target being the JCMT SCUBA-2 450 $\mu\mathrm{m}$ observations of the same field. We reproduce the JCMT SCUBA-2 images with high fidelity using this auto-encoder. This is quantified through the point source fluxes and positions, the completeness, and the purity.- 30 July 2021
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 507 (1) , 1546-1556
Open AccessSuperresolving Herschel imaging: a proof of concept using Deep Neural Networks
Wide-field submillimetre surveys have driven many major advances in galaxy evolution in the past decade, but without extensive follow-up observations the coarse angular resolution of these surveys limits the science exploitation. This has driven the development of various analytical deconvolution methods. In the last half a decade Generative Adversarial Networks have been used to attempt deconvolutions on optical data. Here, we present an auto-encoder with a novel loss function to overcome this problem in the submillimeter wavelength range. This approach is successfully demonstrated on Herschel SPIRE 500 $\mu\mathrm{m}$ COSMOS data, with the superresolving target being the JCMT SCUBA-2 450 $\mu\mathrm{m}$ observations of the same field. We reproduce the JCMT SCUBA-2 images with high fidelity using this auto-encoder. This is quantified through the point source fluxes and positions, the completeness, and the purity. - Open Access
- 29 October 2020
- Published by Association for Computing Machinery (ACM) in Ubiquity
- Vol. 2020 (October) , 1-14
COVID-19 and computation for policy
Governments across the world are formulating and implementing medical, social, economic and other policies to manage the COVID-19 pandemic and protect their citizens. Many governments claim that their policies follow the best available scientific advice. Much of that advice comes from computational modeling. Two of the main types of model are presented: the SIR (Susceptible, Infected, Recovered) model developed by Kermack and McKendrick in the 1920s and the more recent Agent Based Models. The SIR model gives a good intuition of how epidemics spread; including how mass vaccination can contain them. It is less useful than Agent Based Modeling for investigating the effects of policies such as social distancing, self-isolation, wearing facemasks, and test-trace-isolate. Politicians and the public have been perplexed to observe the lack of consensus in the scientific community and there being no single 'best science' to follow. The outcome of computational models depends on the assumptions made and the data used. Different assumptions will lead to different computational outcomes, especially when the available data are so poor. This leads some commentators to argue that the models are wrong and dangerous. Some may be, but computational modeling is one of the few ways available to explore and try to understand the space of possible futures. This lack of certainty means that computational modeling must be seen as just one of many inputs into the political decision making process. Politicians must balance all the competing inputs and make timely decisions based on their conclusions---be they right or wrong. In the same way that democracy is the least worst form of government, computational modeling may be the least worst way of trying to understand the future for policy making.
- 29 October 2020
- Published by Association for Computing Machinery (ACM) in Ubiquity
- Vol. 2020 (October) , 1-14
Open AccessCOVID-19 and computation for policy
Governments across the world are formulating and implementing medical, social, economic and other policies to manage the COVID-19 pandemic and protect their citizens. Many governments claim that their policies follow the best available scientific advice. Much of that advice comes from computational modeling. Two of the main types of model are presented: the SIR (Susceptible, Infected, Recovered) model developed by Kermack and McKendrick in the 1920s and the more recent Agent Based Models. The SIR model gives a good intuition of how epidemics spread; including how mass vaccination can contain them. It is less useful than Agent Based Modeling for investigating the effects of policies such as social distancing, self-isolation, wearing facemasks, and test-trace-isolate. Politicians and the public have been perplexed to observe the lack of consensus in the scientific community and there being no single 'best science' to follow. The outcome of computational models depends on the assumptions made and the data used. Different assumptions will lead to different computational outcomes, especially when the available data are so poor. This leads some commentators to argue that the models are wrong and dangerous. Some may be, but computational modeling is one of the few ways available to explore and try to understand the space of possible futures. This lack of certainty means that computational modeling must be seen as just one of many inputs into the political decision making process. Politicians must balance all the competing inputs and make timely decisions based on their conclusions---be they right or wrong. In the same way that democracy is the least worst form of government, computational modeling may be the least worst way of trying to understand the future for policy making.