2022A&A...664A...4T -
Astronomy and Astrophysics, volume 664A, 4-4 (2022/8-1)
Finding strong gravitational lenses through self-attention. Study based on the Bologna Lens Challenge.
THURUTHIPILLY H., ZADROZNY A., POLLO A. and BIESIADA M.
Abstract (from CDS):
Context. The upcoming large-scale surveys, such as the Rubin Observatory Legacy Survey of Space and Time, are expected to find approximately 105 strong gravitational lenses by analysing data many orders of magnitude larger than those in contemporary astronomical surveys. In this case, non-automated techniques will be highly challenging and time-consuming, if they are possible at all. Aims. We propose a new automated architecture based on the principle of self-attention to find strong gravitational lenses. The advantages of self-attention-based encoder models over convolution neural networks (CNNs) are investigated, and ways to optimise the outcome of encoder models are analysed. Methods. We constructed and trained 21 self-attention-based encoder models and five CNNs to identify gravitational lenses from the Bologna Lens Challenge. Each model was trained separately using 18000 simulated images, cross-validated using 2000 images, and then applied to a test set with 100 000 images. We used four different metrics for evaluation: classification accuracy, the area under the receiver operating characteristic (AUROC) curve, and TPR0 and TPR10 scores (two metrics of evaluation for the Bologna challenge). The performance of self-attention-based encoder models and CNNs participating in the challenge are compared. Results. The encoder models performed better than the CNNs. They were able to surpass the CNN models that participated in the Bologna Lens Challenge by a high margin for the TPR0 and TPR10. In terms of the AUROC, the encoder models with 3 x 106 parameters had equivalent scores to the top CNN model, which had around 23 x 106 parameters. Conclusions. Self-attention-based models have clear advantages compared to simpler CNNs. They perform competitively in comparison to the currently used residual neural networks. Self-attention-based models can identify lensing candidates with a high confidence level and will be able to filter out potential candidates from real data. Moreover, introducing the encoder layers can also tackle the overfitting problem present in the CNNs by acting as effective filters.
NAME Hubble Ultra Deep Field
--
Region defined in the Sky
Origin of the objects types :
(Ref) Object type as listed in the reference "Ref"
(acronym) Object type linked to the acronym according to the original reference
() Anterior to 2007, before we can link the objet type to a reference, or given by the CDS team in some particular cases
Other object types:
reg
()
Syntax of coordinates is : "ra dec (wtype) [error ellipse] quality bibcode" :
ra dec : right ascension and declination (unit and frame defined according to your Output Options)
Grey values are increasing the original precision due to the computation of frame transformations
(wtype) : wavelength class for the origin of the coordinates (Rad, mm, IR, Optical, UV, Xray, Gam)
[error ellipse] : measurement uncertainty, on (ra,dec) if the positional angle is 90 degrees, on (majaxis,minaxis) otherwise (in mas at defined epoch in the original catalogue),
position angle (in degrees North celestial pole to East)
quality : flag of quality
E ≥ 10"
D : 1-10" (and some old data)
C : 0.1-1"
B : 0.01-0.1" + 2MASS, Tyc
A : VLBI, Hipparcos
bibcode : bibcode of the coordinates reference
ICRS
coord.
(ep=J2000) :
03 32 39.0 -27 47 29
[
]
D
~
Syntax of coordinates is : "ra dec (wtype) [error ellipse] quality bibcode" :
ra dec : right ascension and declination (unit and frame defined according to your Output Options)
Grey values are increasing the original precision due to the computation of frame transformations
(wtype) : wavelength class for the origin of the coordinates (Rad, mm, IR, Optical, UV, Xray, Gam)
[error ellipse] : measurement uncertainty, on (ra,dec) if the positional angle is 90 degrees, on (majaxis,minaxis) otherwise (in mas at defined epoch in the original catalogue),
position angle (in degrees North celestial pole to East)
quality : flag of quality
E ≥ 10"
D : 1-10" (and some old data)
C : 0.1-1"
B : 0.01-0.1" + 2MASS, Tyc
A : VLBI, Hipparcos
bibcode : bibcode of the coordinates reference
FK4
coord.
(ep=B1950 eq=1950) :
03 30 33.4 -27 57 34
[
]
Syntax of coordinates is : "ra dec (wtype) [error ellipse] quality bibcode" :
ra dec : right ascension and declination (unit and frame defined according to your Output Options)
Grey values are increasing the original precision due to the computation of frame transformations
(wtype) : wavelength class for the origin of the coordinates (Rad, mm, IR, Optical, UV, Xray, Gam)
[error ellipse] : measurement uncertainty, on (ra,dec) if the positional angle is 90 degrees, on (majaxis,minaxis) otherwise (in mas at defined epoch in the original catalogue),
position angle (in degrees North celestial pole to East)
quality : flag of quality
E ≥ 10"
D : 1-10" (and some old data)
C : 0.1-1"
B : 0.01-0.1" + 2MASS, Tyc
A : VLBI, Hipparcos
bibcode : bibcode of the coordinates reference
Gal
coord.
(ep=J2000) :
223.5548 -54.3948
[
]
', {sourceSize:12, color:'#30a090'})); aladin.on('objectClicked', function(object) { var objName=object.data.MAIN_ID; aladin.showPopup(object.ra,object.dec,'',''+ objName+''); });" title="Show Simbad objects">
Overlay
points in this preview
All
(CDSPortal)
Send to
The link on the acronym of the identifiers give access to the
information for this acronym in the dictionary of nomenclature.
Identifiers (3) :
An access of full data is available using the icon Vizier near the identifier of the catalogue
References (1666 between 1850 and 2024) (Total 1666)
Simbad bibliographic survey began in 1850 for stars (at least bright stars) and in 1983 for all other objects (outside the solar system).
Follow
new references on this object
Annotations :
Annotations allow a user to add a note or report an error concerning the astronomical object and its data. It requires registration to post a note. See description . Please, have a look at Best practices. The list of all annotations to SIMBAD objects can be found here .
To bookmark this query, right click on this link: simbad:objects in 2022A&A...664A...4T and select 'bookmark this link' or equivalent in the popup menu