Identification of extragalactic radio sources in the VISTA survey of the Magellanic Clouds using an unsupervised machine learning algorithm

arXiv:2611.15880 · astro-ph.GA, astro-ph.IM · Submitted 2026-11-11 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: We're looking at the paper: "Identification of extragalactic radio sources in the VISTA survey of the Magellanic Clouds using an unsupervised machine learning algorithm".

Jocelyn: Here's the gist of the paper. The paper identifies extragalactic radio sources in the Magellanic Clouds using an unsupervised machine learning algorithm to separate stellar and extragalactic populations The combination of unsupervised machine learning with photometry from optical to radio has proven adept at separating the stellar from the extragalactic, and the AGN from the galaxies Data Acquisition and Feature Engineering The study utilizes a multi-wavelength dataset combining observations from several surveys to create features for source classification The photometric surveys selected aim to provide large coverage of the VISTA Magellanic Clouds and the ASKAP SMC and LMC footprints, focusing on the synergy of near-IR and radio for detecting extragalactic sources in stellar dense fields Features included in the t-SNE algorithm are derived from various data, such as optical to infrared photometry, X-ray flux, redshift from spectroscopic observations, variability information from the VMC survey, and proper motions in RA and DEC The creation of these features involves calculating colors by subtracting each feature from all the other features of the same survey to create them Unsupervised Learning Methodology The core method employed is the t-SNE algorithm, a non-linear dimensionality reduction algorithm used to visualize high-dimensional datasets in two dimensions t-SNE minimizes the divergence between a probability distribution that measures pairwise similarities of the high-dimensional data and a probability distribution that measures pairwise similarities of the low-dimensional points in the embedding, preserving local structure The algorithm is unsupervised, meaning it learns from all the data to structure and separate sources without requiring a training set A perplexity parameter of around 50 was chosen as it showed well-structured clustering for the results Clustering Results and Source Identification The application of t-SNE reveals distinct clusters corresponding to different source types, with some overlap between classes Extragalactic, Magellanic and foreground objects separate from each other into noticeable clusters, with some overlap AGN tend to be found across the centres of the map, at one end of which a population of blazars can be found The Gaia BP/RP excess and VISTA sharpness values trace how extended an object is, causing a gradient in redshift across the extragalactic population on the t-SNE map The AllWISE detections that were upper limits were not removed when collating the photometry, and these sources are clearly separated from those that have good detections in all bands Validation and Interpretation The results are validated by comparing the t-SNE map clustering to other classification methods, such as Gaia classifications and supervised machine learning (PRF) The comparison between the unsupervised t-SNE and supervised PRF shows that the two methods separate the different types of sources well Follow-up optical spectroscopy of the blazar candidate sequence with SALT has revealed some strong blazar candidates can be found with the t-SNE map, although further information is needed alongside the t-SNE to choose the more likely blazar candidates The majority of X-ray detected sources are found where AGN are predicted to lie, as expected, including a blazar candidate sequence The study demonstrates broad agreement between different classification schemes – t-SNE, PRF, Gaia and spectroscopic – which can complement one another to obtain more complete samples Future work could involve removing sources with only upper limits in WISE data and seeing how this affects the clustering It is possible that these upper limits helped separate the extragalactic from the stellar, so another avenue of exploration would be to put back in the sources that had missing data from any band and provide an upper limit for the missing data The similarity between the maps suggests that the two fields contain broadly comparable classes of radio-detected sources Table A3 Table A3 lists the serendipitous spectral classifications, showing how different source types are distributed across the t-SNE maps The PRF classifies the SMC sources (with probabilities ∼0.22–0.38) SMC tSNE b1, b2, b3 and b4 as AGN, SMC tSNE b5, b6 and b7 as galaxies The PRF classifies the LMC source (with probabilities ∼0.58–0.97) LMC tSNE b1 as an H ii/YSO, which matches with it being an emission line object Table A2 Table A2 provides the source classifications for each target, detailing the known source classification and this work classification For example, SMC tSNE b4 is classified as an RGB star by this work LMC tSNE b4 is classified as a Blazar by this work The LMC tSNE b9 is classified as a Blazar by this work Table A3 Table A3 lists the serendipitous spectral classifications, showing how different source types are distributed across the t-SNE maps The LMC tSNE b9 is classified as an A-type star by this work Figure 3 Figure 3 compares the distribution of proper motions of the sources in the t-SNE maps of the SMC (left) and LMC (right) The highest PM values are where high proper-motion stars are located, and the lowest proper motions are where AGN are located Figure 4 Figure 4 compares the distribution of the median value of Gaia BP/RP excess of the sources in the t-SNE maps of the SMC (left) and LMC (right) Galaxies can be found where the greatest difference between Gaia G and the GRP and GBP bands is, as expected Figure 5 Figure 5 compares the distribution of the median value of VMC PSF colour J−Ks (mag) of the sources in the t-SNE maps of the SMC (left) and LMC (right) The extragalactic regions of the map have redder J−Ks colours Figure 6 Figure 6 compares the distribution of the median value of VMC PSF J-band sharpness value of the sources in the t-SNE maps of the SMC (left) and LMC (right) The more extended objects line up with the low-z galaxies Figure 7 Figure 7 compares the distribution of the median value of AllWISE colour W1−W2 of the sources in the t-SNE maps of the SMC (left) and LMC (right) Known AGN clearly stand out in this AllWISE colour, showing that this WISE colour is good for selecting AGN Figure 9 Figure 9 compares the distribution of redshift in the t-SNE maps of the SMC (left) and LMC (right) There is a redshift gradient from top to bottom for the SMC and right to left for the LMC The lowest redshifts match with the most extended sources, which are the low-z galaxies Figure 8 Figure 8 compares the distribution of the median value of ASKAP 960/888 MHz flux density in the t-SNE maps of the SMC (left) and LMC (right) The area of the brightest radio sources lines up with the area of known blazars, as expected Figure 2 Figure 2 shows the t-SNE map of the SMC (left) and LMC (right) where known, spectroscopically observed AGN are highlighted in cyan and SAGE0536AGN and SAGE0534AGN have been highlighted in red The rectangular regions show the general area of specific classifications of sources Figure 4 Figure 4 shows the comparison of the distribution of the median value of Gaia BP/RP excess across the t-SNE maps The rectangular regions are the same as in Figure 2 Figure 5 Figure 5 shows the comparison of the distribution of the median value of VMC PSF colour J−Ks (mag) across the t-SNE maps The extragalactic regions of the map have redder J−Ks colours Figure 6 Figure 6 shows the comparison of the distribution of the median value of VMC PSF J-band sharpness value across the t-SNE maps The more extended objects line up with the low-z galaxies Figure 7 Figure 7 shows the comparison of the distribution of the median value of AllWISE colour W1−W2 across the t-SNE maps Known AGN clearly stand out in this AllWISE colour, showing that this WISE colour is good for selecting AGN Figure 8 Figure 8 shows the comparison of the distribution of the median value of ASKAP 960/888 MHz flux density across the t-SNE maps The area of the brightest radio sources lines up with the area of known blazars, as expected Figure 3 Figure 3 shows the comparison of the distribution of proper motions of the sources in the t-SNE maps of the SMC (left) and LMC (right) The highest PM values are where high proper-motion stars are located, and the lowest PM values are where AGN are located Figure 10 Figure 10 shows how the median value of VMC probability of a source being variable in the Ks band varies across the tSNE maps In the areas where AGN/blazars reside, the variability probability is generally higher (more yellow) Figure 3 Figure 3 shows the comparison of the distribution of proper motions of the sources in the t-SNE maps of the SMC (left) and LMC (right) High proper-motion stars are located in the isolated bottom and top left islands of the SMC and LMC t-SNE maps, respectively Figure 4 Figure 4 shows the comparison of the distribution of the median value of Gaia BP/RP excess across the t-SNE maps High values of Gaia BP/RP excess indicate a large difference between G flux and the average of integrated flux values, which is expected for extended objects Figure 7 Figure 7 shows the comparison of the distribution of the median value of AllWISE colour W1−W2 across the t-SNE maps The highest values coincide with the known AGN, showing that this WISE colour is good for selecting AGN Figure 9 Figure 9 compares the distribution of redshift in the t-SNE maps of the SMC (left) and LMC (right) The lowest redshifts match with the most extended sources, which are the low-z galaxies Figure 4 Figure 4 shows how Gaia E(BP/RP) varies across the t-SNE maps Galaxies can be found where the greatest difference between Gaia G and the GRP and GBP bands is, as expected Figure 5 Figure 5 shows how the colour J−Ks varies across class positions on the map The extragalactic regions of the map have redder J−Ks colours Figure 7 Figure 7 shows how the WISE colour W1−W2 varies across class positions on the map The highest values coincide with the known AGN, showing that this WISE colour is good for selecting AGN Figure 4

Title and authors: Vera: Now, let’s get into the specific details of what they found in this paper about its summary. They are talking about how they used this combination of optical to radio photometry to sort out stellar populations and AGN from galaxies.

Jocelyn: So, basically, the paper is showing that combining near-infrared and radio data is a powerful tool for identifying active galactic nuclei because the dusty torus in those sources has a distinct bump-like shape in their spectral energy distribution.

Subrahmanyan: That’s a key piece of physics. The infrared photometry helps isolate these sources in color-color diagrams, like the one Werner et al. two thousand four discussed, and the radio detection itself is also a strong sign that we are observing an AGN because star formation or stellar sources don't usually exhibit those flat radio spectral slopes that AGN do <ref:2611.15880#pg1>.

Vera: Right. But they also look at how extended these objects are using Gaia BP over RP excess and VISTA sharpness values, which gives them clues about the redshift gradient across the extragalactic population on their t-SNE map.

Jocelyn: And they’re showing that the AllWISE color W1 minus W2 is particularly useful for selecting AGN because the highest values coincide with known AGN on their maps. It’s a very direct indicator.

Subrahmanyan: That connection between physical characteristics—like how extended an object is, traced by sharpness or the color difference between Gaia bands—and the derived redshift gradient is what links structure to cosmology here. If we can map that out, it helps constrain models of how these systems evolve.

Vera: They’ve also mapped out redshift directly on their t-SNE maps, showing a clear gradient from top to bottom for the SMC and right to left for the LMC. The lowest redshifts match with the most extended sources, which are those low-z galaxies.

Jocelyn: That gradient is telling us something about how these Magellanic Clouds are distributed in space, not just within themselves. It suggests that their structure isn't uniform across the entire field of view.

Subrahmanyan: And they used the ASKAP flux density data to line up the area of the brightest radio sources with known blazars, which confirms that this radio data is tracking those specific energetic phenomena we are interested in studying.

The paper's summary: Vera: So, moving on to what they suggest as improvements for this approach, they point out a few ways the methodology can be strengthened. They want to enhance source classification capabilities using unsupervised machine learning by combining multi-wavelength photometry from optical to radio.

Jocelyn: And they specifically highlight that feature selection can improve identification accuracy by using features like J minus Ks color and W1 minus W2 color, noting that the W1 minus W2 color is particularly effective across class positions on the map.

Subrahmanyan: From a theoretical view, incorporating Gaia proper motions and AllWISE quality flags into machine learning algorithms could help identify false matches upfront, which is a technique that can be unbiased from human selection before you even start classifying.

Vera: They suggest refining the classification of AGN candidates by combining the unsupervised t-SNE results with supervised Probabilistic Random Forest classifiers, because they found that comparing these two methods separates the sources well.

Jocelyn: And they also propose a more robust identification of blazar candidates by assessing variability probability in near-infrared bands, because they observed that the highest probabilities for a source being variable are found amongst the AGN area.

Subrahmanyan: That variability check is interesting because it connects to the physics of AGN activity; if you see high variability probability in those regions, it strongly suggests we’re looking at active sources. It provides an extra layer of confirmation beyond just spatial clustering on the map.

Vera: So, these improvements focus on adding more discriminating features and cross-validation between unsupervised and supervised methods to make the identification more reliable overall.

Jocelyn: It seems like they are trying to build a system that is less reliant on any single piece of data, pushing toward a more holistic classification strategy.

The paper's improvements: Vera: So, wrapping up this paper on "Identification of extragalactic radio sources in the VISTA survey of the Magellanic Clouds using an unsupervised machine learning algorithm," it shows broad agreement across different classification schemes. The t-SNE map clustering aligns well with Gaia classifications and supervised PRF methods.

Jocelyn: And they also found that follow-up optical spectroscopy of the blazar candidate sequence with SALT has revealed some strong blazar candidates on the map, although they stress that further information is needed to choose the most likely ones.

Subrahmanyan: The overall implication is that these different classification schemes—t-SNE, PRF, Gaia classifications, and spectroscopic data—can complement each other to obtain more complete samples of sources in these regions.

Vera: It means we aren't just relying on one method; we have a toolkit here to build more reliable catalogs for the Magellanic Clouds and beyond. The work demonstrates that these different fields can indeed provide complementary information.

Jocelyn: I think this suggests that the similarity between these maps points to broadly comparable classes of radio-detected sources, even if the underlying classification methods are different.

Subrahmanyan: So, in short, this study uses the t-SNE map as a powerful visualization tool to show how different physical properties correlate with source types. The paper is about identifying extragalactic radio sources in the VISTA survey of the Magellanic Clouds using an unsupervised machine learning algorithm.

Vera: That’s right. We’re looking at how this study uses that algorithm to separate stellar and extragalactic populations, and AGN from galaxies using multi-wavelength data from optical to radio.

Jocelyn: And they are setting up a future direction by suggesting they look at removing sources with only upper limits in WISE data and see how that affects the clustering.

Subrahmanyan: That’s a good next step because it tests if those upper limits actually help separate the extragalactic from the stellar, which is what we want to test for structure in this field.

Vera: It seems like they’re aiming to build a more complete sample by incorporating more complex data constraints into their classification process. That's where we go next.

Conclusion: Vera: So we've been looking at how they used that unsupervised machine learning algorithm to separate stellar from extragalactic sources and AGN from galaxies in those Magellanic Clouds using optical to radio photometry.

Jocelyn: It really shows how combining those different wavelengths, near-infrared and radio, proves adept at sorting out the stellar stuff from the extragalactic population, and also AGN from the galaxies.

Subrahmanyan: That’s interesting because it links physical characteristics—like how extended an object is—to a redshift gradient across that t-SNE map.

Vera: Yeah, and they validated their t-SNE clusters by comparing them to Gaia classifications and a supervised probabilistic random forest classifier, showing both methods separate the source types well.

Jocelyn: And they even found strong blazar candidates in those maps using follow-up optical spectroscopy with SALT, which is a good sign.

Subrahmanyan: It confirms what we suspected about the AGN being where X-ray sources are predicted to lie, including that blazar candidate sequence.

Vera: The study ultimately shows broad agreement across all these classification schemes—t-SNE, PRF, Gaia and spectroscopic data—which is what we hoped for.

Jocelyn: It's a solid piece of work because it suggests that these different methods can actually complement each other to get a more complete picture of the sources.

Subrahmanyan: The paper, "Identification of extragalactic radio sources in the VISTA survey of the Magellanic Clouds using an unsupervised machine learning algorithm," confirms that these radio-detected sources are broadly comparable classes, which is important for our cosmological models.

Vera: Exactly. They also pointed out some future work, like removing sources with only upper limits in WISE data to see how that affects the clustering, which could separate the extragalactic from the stellar population better.

Jocelyn: That’s a good avenue to explore because those upper limits might have been key in separating them initially.

Subrahmanyan: I think that's a crucial test for constraining structure formation models, and it gives us more control over what we include in the sample.

Vera: Well, that wraps up our look at this paper; it’s a great example of how unsupervised learning can be useful when you combine so many different data types.

Jocelyn: It leaves us with a clear idea that looking at the structure of these radio sources in the Magellanic Clouds gives us some interesting clues about what's happening out there.

Subrahmanyan: And it opens up more avenues for how we use these multi-wavelength datasets to map cosmic structure, which is where the real big picture lies.

Clara M. Pennock, Jacco Th. van Loon, Mohammed Iddrisu Nlowie, Elizabeth Joy Leese, Maja Nowaczyk, Heinz Andernach, Duncan Farrah, Valentin D. Ivanov, Michał J. Michałowski, Joana M. Oliveira

Institute for Astronomy, University of Edinburgh · School of Engineering and Physical Sciences, Lennard-Jones Laboratories, Keele University School of Engineering and Physical Sciences at Keele University Department de Astronomía, Universidad de Guanajuato Thüringer Landessternwarte Thüringer Landessternwarte

astro-ph.GA, astro-ph.IM

Submitted: 2026-11-11

Updated: 2026-11-11

Comments: MNRAS accepted

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 85/100

The gist: The paper identifies extragalactic radio sources in the Magellanic Clouds using an unsupervised machine learning algorithm to separate stellar and extragalactic populations The combination of…

Key concepts

t-SNE Algorithm
A non-linear dimensionality reduction algorithm used to visualize high-dimensional data in two dimensions. It groups similar data points together based on their relationships in the original complex feature space, helping researchers see distinct clusters of different source types without needing prior labels.
Multi-wavelength Photometry
Using observations from various wavelengths, including optical, infrared (VISTA), X-ray, and radio surveys. Combining data from these different parts of the spectrum allows the algorithm to capture unique characteristics of sources like stars versus distant galaxies or active galactic nuclei.
Unsupervised Learning
A machine learning approach where the algorithm learns to find patterns and structure in data without being explicitly told what those patterns should look like beforehand. In this study, it was used to automatically discover natural groupings among the observed sources.

Terminology

Summary

The paper identifies extragalactic radio sources in the Magellanic Clouds using an unsupervised machine learning algorithm to separate stellar and extragalactic populations The combination of unsupervised machine learning with photometry from optical to radio has proven adept at separating the stellar from the extragalactic, and the AGN from the galaxies

Data Acquisition and Feature Engineering

The study utilizes a multi-wavelength dataset combining observations from several surveys to create features for source classification The photometric surveys selected aim to provide large coverage of the VISTA Magellanic Clouds and the ASKAP SMC and LMC footprints, focusing on the synergy of near-IR and radio for detecting extragalactic sources in stellar dense fields Features included in the t-SNE algorithm are derived from various data, such as optical to infrared photometry, X-ray flux, redshift from spectroscopic observations, variability information from the VMC survey, and proper motions in RA and DEC The creation of these features involves calculating colors by subtracting each feature from all the other features of the same survey to create them

Unsupervised Learning Methodology

The core method employed is the t-SNE algorithm, a non-linear dimensionality reduction algorithm used to visualize high-dimensional datasets in two dimensions t-SNE minimizes the divergence between a probability distribution that measures pairwise similarities of the high-dimensional data and a probability distribution that measures pairwise similarities of the low-dimensional points in the embedding, preserving local structure The algorithm is unsupervised, meaning it learns from all the data to structure and separate sources without requiring a training set A perplexity parameter of around 50 was chosen as it showed well-structured clustering for the results

Clustering Results and Source Identification

The application of t-SNE reveals distinct clusters corresponding to different source types, with some overlap between classes Extragalactic, Magellanic and foreground objects separate from each other into noticeable clusters, with some overlap AGN tend to be found across the centres of the map, at one end of which a population of blazars can be found The Gaia BP/RP excess and VISTA sharpness values trace how extended an object is, causing a gradient in redshift across the extragalactic population on the t-SNE map The AllWISE detections that were upper limits were not removed when collating the photometry, and these sources are clearly separated from those that have good detections in all bands

Validation and Interpretation

The results are validated by comparing the t-SNE map clustering to other classification methods, such as Gaia classifications and supervised machine learning (PRF) The comparison between the unsupervised t-SNE and supervised PRF shows that the two methods separate the different types of sources well Follow-up optical spectroscopy of the blazar candidate sequence with SALT has revealed some strong blazar candidates can be found with the t-SNE map, although further information is needed alongside the t-SNE to choose the more likely blazar candidates The majority of X-ray detected sources are found where AGN are predicted to lie, as expected, including a blazar candidate sequence

Conclusion and Future Directions

The study demonstrates broad agreement between different classification schemes – t-SNE, PRF, Gaia and spectroscopic – which can complement one another to obtain more complete samples Future work could involve removing sources with only upper limits in WISE data and seeing how this affects the clustering It is possible that these upper limits helped separate the extragalactic from the stellar, so another avenue of exploration would be to put back in the sources that had missing data from any band and provide an upper limit for the missing data The similarity between the maps suggests that the two fields contain broadly comparable classes of radio-detected sources

Table A3

Table A3 lists the serendipitous spectral classifications, showing how different source types are distributed across the t-SNE maps The PRF classifies the SMC sources (with probabilities ∼0.22–0.38) SMC tSNE b1, b2, b3 and b4 as AGN, SMC tSNE b5, b6 and b7 as galaxies The PRF classifies the LMC source (with probabilities ∼0. p. 5

Where the paper says it

The co-ordinates of the SMC and LMC sample investigated in this work along with the two dimensions that resulted from the tSNE algorithm, 𝑥 and 𝑦 (e. as seen in Figure 2), and the PRF classifications and their probabilities from Pennock et al.

58–0.97) LMC tSNE b1 as an H ii/YSO, which matches with it being an emission line object

Table A2

Table A2 provides the source classifications for each target, detailing the known source classification and this work classification For example, SMC tSNE b4 is classified as an RGB star by this work LMC tSNE b4 is classified as a Blazar by this work The LMC tSNE b9 is classified as a Blazar by this work

Table A3 lists the serendipitous spectral classifications, showing how different source types are distributed across the t-SNE maps The LMC tSNE b9 is classified as an A-type star by this work

Figure 3

Figure 3 compares the distribution of proper motions of the sources in the t-SNE maps of the SMC (left) and LMC (right) The highest PM values are where high proper-motion stars are located, and the lowest proper motions are where AGN are located

Figure 4

Figure 4 compares the distribution of the median value of Gaia BP/RP excess of the sources in the t-SNE maps of the SMC (left) and LMC (right) Galaxies can be found where the greatest difference between Gaia G and the GRP and GBP bands is, as expected

Figure 5

Figure 5 compares the distribution of the median value of VMC PSF colour J−Ks (mag) of the sources in the t-SNE maps of the SMC (left) and LMC (right) The extragalactic regions of the map have redder J−Ks colours

Figure 6

Figure 6 compares the distribution of the median value of VMC PSF J-band sharpness value of the sources in the t-SNE maps of the SMC (left) and LMC (right) The more extended objects line up with the low-z galaxies

Figure 7

Figure 7 compares the distribution of the median value of AllWISE colour W1−W2 of the sources in the t-SNE maps of the SMC (left) and LMC (right) Known AGN clearly stand out in this AllWISE colour, showing that this WISE colour is good for selecting AGN

Figure 9

Figure 9 compares the distribution of redshift in the t-SNE maps of the SMC (left) and LMC (right) There is a redshift gradient from top to bottom for the SMC and right to left for the LMC The lowest redshifts match with the most extended sources, which are the low-z galaxies

Figure 8

Figure 8 compares the distribution of the median value of ASKAP 960/888 MHz flux density in the t-SNE maps of the SMC (left) and LMC (right) The area of the brightest radio sources lines up with the area of known blazars, as expected

Figure 2

Figure 2 shows the t-SNE map of the SMC (left) and LMC (right) where known, spectroscopically observed AGN are highlighted in cyan and SAGE0536AGN and SAGE0534AGN have been highlighted in red The rectangular regions show the general area of specific classifications of sources

Figure 4 shows the comparison of the distribution of the median value of Gaia BP/RP excess across the t-SNE maps The rectangular regions are the same as in Figure 2

Figure 5 shows the comparison of the distribution of the median value of VMC PSF colour J−Ks (mag) across the t-SNE maps The extragalactic regions of the map have redder J−Ks colours

Figure 6 shows the comparison of the distribution of the median value of VMC PSF J-band sharpness value across the t-SNE maps The more extended objects line up with the low-z galaxies

Figure 7 shows the comparison of the distribution of the median value of AllWISE colour W1−W2 across the t-SNE maps Known AGN clearly stand out in this AllWISE colour, showing that this WISE colour is good for selecting AGN

Figure 8 shows the comparison of the distribution of the median value of ASKAP 960/888 MHz flux density across the t-SNE maps The area of the brightest radio sources lines up with the area of known blazars, as expected

Figure 3 shows the comparison of the distribution of proper motions of the sources in the t-SNE maps of the SMC (left) and LMC (right) The highest PM values are where high proper-motion stars are located, and the lowest PM values are where AGN are located

Figure 10

Figure 10 shows how the median value of VMC probability of a source being variable in the Ks band varies across the tSNE maps In the areas where AGN/blazars reside, the variability probability is generally higher (more yellow)

Figure 3 shows the comparison of the distribution of proper motions of the sources in the t-SNE maps of the SMC (left) and LMC (right) High proper-motion stars are located in the isolated bottom and top left islands of the SMC and LMC t-SNE maps, respectively

Figure 4 shows the comparison of the distribution of the median value of Gaia BP/RP excess across the t-SNE maps High values of Gaia BP/RP excess indicate a large difference between G flux and the average of integrated flux values, which is expected for extended objects

Figure 7 shows the comparison of the distribution of the median value of AllWISE colour W1−W2 across the t-SNE maps The highest values coincide with the known AGN, showing that this WISE colour is good for selecting AGN

Figure 9 compares the distribution of redshift in the t-SNE maps of the SMC (left) and LMC (right) The lowest redshifts match with the most extended sources, which are the low-z galaxies

Figure 4 shows how Gaia E(BP/RP) varies across the t-SNE maps Galaxies can be found where the greatest difference between Gaia G and the GRP and GBP bands is, as expected

Figure 5 shows how the colour J−Ks varies across class positions on the map The extragalactic regions of the map have redder J−Ks colours

Figure 7 shows how the WISE colour W1−W2 varies across class positions on the map The highest values coincide with the known AGN, showing that this WISE colour is good for selecting AGN

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Enhanced source classification capabilities using unsupervised machine learning can separate stellar from extragalactic sources and AGN from galaxies by combining multi-wavelength photometry from optical to radio, as shown in This work shows that the combination of unsupervised machine learning with photometry from optical to radio has proven adept at separating the stellar from the extragalactic, and the AGN from the galaxies.

  2. Improved source identification accuracy through feature selection can be achieved by using features such as J minus Ks colour and W1 minus W2 colour, which The WISE parameter that shows the greatest difference across class positions on the map is the W1−W2 colour (Figure 7).

  3. Better distinction between source types can be made by incorporating Gaia proper motions and AllWISE quality flags into machine learning algorithms to identify false matches, as this technique can identify groupings that are unbiased from human selection upfront.

  4. Refined classification of AGN candidates can be achieved by combining unsupervised t-SNE results with supervised Probabilistic Random Forest (PRF) classifiers, as the comparison Overall, the comparison between the unsupervised t-SNE and supervised PRF shows that the two methods separate the different types of sources well.

  5. More robust identification of blazar candidates can be performed by assessing variability probability in near-infrared bands, as The highest probabilities for a source being variable are found amongst the AGN area and "In Figure 10 shows how the median value of variability probability in the VISTA Ks band varies across the tSNE maps. From this we see that in the areas where AGN/blazars reside the variability probability is generally higher (more yellow)."

Abstract

The Magellanic Clouds benefit from a wealth of data across the electromagnetic spectrum, but extracting extragalactic sources from these crowded fields is a less than straightforward endeavour. We used an unsupervised machine learning algorithm (t-distributed stochastic neighbourhood embedding; t-SNE) to classify 6,652 and 1,356 sources in the Large and Small Magellanic Clouds, respectively, that were detected in the VISTA Survey of the Magellanic Clouds (VMC) and the Evolutionary Map of the Universe (EMU) survey with the Australian Square Kilometre Array Pathfinder (ASKAP). Multi-wavelength photometry and colours from optical to radio were used as features to cluster similar sources together in 2D space on a t-SNE map. We used previously classified sources to separate the clusters into five classes: AGN, low-redshift galaxies, blazars, dusty stellar objects (e.g. young stellar objects, planetary nebulæ etc) and high proper-motion stars. Previous photometric classifications from the literature match with specific clusters identified, as well as provide insight on areas of the t-SNE map for which there are no previous classifications. This work shows that the combination of unsupervised machine learning with photometry from optical to radio has proven adept at separating the stellar from the extragalactic, and the AGN from the galaxies. Our associated new spectroscopy with SALT includes new discoveries of AGN up to redshift 3.3 and Magellanic emission-line objects.

Sources

Related papers