QSO photometric redshifts using machine learning and neural networks

Curran, S. J.; Moss, J. P.; Perrott, Y. C.

doi:10.1093/mnras/stab485

Cited by 25 publications

(45 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The best results are obtained when the photometry from nine bands, spanning the farultraviolet to near-infrared (F U V, N U V, u, g, r, i, z, W 1, W 2), are used. Scraping the photometry from the NASA/IPAC Extragalactic Database, the Wide-Field Infrared Survey Explorer, the Two Micron All Sky Survey (Skrutskie et al 2006) and Galaxy Evolution Explorer (Bianchi et al 2017) databases, as described in Curran et al (2021), we find the photometry to be limited to ν 20 GHz apart from one X-ray measurement. The lack of photometry, therefore, does not permit a photometric redshift determination.…”

Section: The Absorption Towards Pks 1657-298mentioning

confidence: 95%

“…As described in Sect. 1, the latter can be obtained from a neural network trained on SDSS QSOs and then confidently applied to a radio selected source (Curran et al 2021). The best results are obtained when the photometry from nine bands, spanning the farultraviolet to near-infrared (F U V, N U V, u, g, r, i, z, W 1, W 2), are used.…”

Section: The Absorption Towards Pks 1657-298mentioning

confidence: 99%

“…Training on optically selected samples (the Sloan Digital Sky Survey, SDSS), shows promising results with the validation giving accurate redshift predictions for quasi-stellar objects (QSOs, Laurino et al 2011;Brescia et al 2013;D'Isanto & Polsterer 2018;Pasquet-Itam & Pasquet 2018;Beck et al 2021). Training on SDSS QSOs can also provide a model which accurately predicts redshifts for radio selected sources (quasars, Curran et al 2021). The model, however, requires measurements in nine photometric bands, spanning from the nearinfrared to far-ultraviolet, which may not necessarily be available.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Intervening or associated? Machine learning classification of redshifted H I 21-cm absorption

Curran

2021

Preprint

View full text Add to dashboard Cite

In a previous paper we presented the results of applying machine learning to classify whether an H I 21-cm absorption spectrum arises in a source intervening the sight-line to a more distant radio source or within the host of the radio source itself. This is usually determined from an optical spectrum giving the source redshift. However, not only will this be impractical for the large number of sources expected to be detected with the Square Kilometre Array, but bright optical sources are the most ultra-violet luminous at high redshift and so bias against the detection of cool, neutral gas. Adding another 44, mostly newly detected absorbers, to the previous sample of 92, we test four different machine learning algorithms, again using the line properties (width, depth and number of Gaussian fits) as features. Of these algorithms, three gave a some improvement over the previous sample, with a logistic regression model giving the best results. This suggests that the inclusion of further training data, as new absorbers are detected, will further increase the prediction accuracy above the current ≈ 80%. We use the logistic regression model to classify the z abs = 0.42 absorption towards PKS 1657-298 and find this to be associated, which is consistent with a previous study which determined z em ≈ 0.42 from the K-band magnitude-redshift relation.

show abstract

Section: The Absorption Towards Pks 1657-298mentioning

confidence: 95%

Section: The Absorption Towards Pks 1657-298mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Intervening or associated? Machine learning classification of redshifted H I 21-cm absorption

Curran

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…is due to a single object. For this we use the photometry compiled by Curran et al (2021) for the first 100 337 QSOs with accurate spectroscopic redshifts (δz/z < 0.01) of the Sloan Digital Sky Survey (SDSS) Data Release 12 (DR12, Alam et al 2015). With a central wavelength of λ = 551 nm, the V -band is located between the SDSS r (λ = 623 nm) and g (λ = 477 nm) bands.…”

Section: Reddening Of 4c +0519mentioning

confidence: 99%

On the scarcity of redshifted OH and millimetre-band molecular absorption

Curran

2021

Preprint

View full text Add to dashboard Cite

Despite much searching, redshifted decimetre and millimetre-band absorption by molecular gas remains very rare, limited to just six systems at z abs > ∼ 0.05. Detection of these transitions can yield precise diagnostics of the conditions of the star forming gas in the earlier Universe, the hydroxyl (OH) radical being of particular interest as in the λ = 18 cm ground state there are four different transitions located close to H I 21-cm and thus detectable with the Square Kilometre Array and its pathfinders. The four transitions of OH have very different dependences on the fundamental constants, thus having much potential in testing for any evolution in these over large look-back times. By collating the photometry in a uniform manner, we confirm our previous hypothesis that the normalised OH absorption strength is correlated with the optical-near-infrared red colour of the sight-line. Applying this to the published searches, we find that all, but one (J0414+054), have simply not been searched sufficiently deeply. We suggest that this is due to the standard selection of sources with reliable optical redshifts introducing a bias against those with enough dust with which to shield the molecular gas. For the single source searched to sufficient depth, we have reason to suspect that the high degree of reddening arises from another system along the sight-line, thus not being inconsistent with our hypothesis. We also show that the same optical redshift bias can account for the scarcity of millimetre-band absorption.

show abstract

“…The "+" symbols in the foreground represent the galaxy's photometric measurements in each band, and are coloured to match the corresponding filter in the background. (Ball et al, 2007(Ball et al, , 2008Oyaizu et al, 2008;Zhang et al, 2013;Kügler et al, 2015;Cavuoti et al, 2017;Luken et al, 2019Luken et al, , 2021, Random Forest (RF; Cavuoti et al, 2012Cavuoti et al, , 2015Hoyle, 2016;Sadeh et al, 2016;Cavuoti et al, 2017;Pasquet-Itam and Pasquet, 2018), and neural networks (Firth et al, 2003;Tagliaferri et al, 2003;Collister and Lahav, 2004;Brodwin et al, 2006;Oyaizu et al, 2008;Hoyle, 2016;Sadeh et al, 2016;Curran, 2020;Curran et al, 2021) being among the more widely used algorithms. Some recent studies utilise Gaussian Process (GP; Duncan et al, 2018a,b;Duncan et al, 2021), and deep learning using the original images at different wavelengths, as opposed to the photometry extracted from the image (D'Isanto and Polsterer, 2018).…”

Section: Introductionmentioning

confidence: 99%

Estimating Galaxy Redshift in Radio-Selected Datasets using Machine Learning

Luken¹,

Norris²,

Park³

et al. 2022

Preprint

View full text Add to dashboard Cite

All-sky radio surveys are set to revolutionise the field with new discoveries. However, the vast majority of the tens of millions of radio galaxies won't have the spectroscopic redshift measurements required for a large number of science cases. Here, we evaluate techniques for estimating redshifts of galaxies from a radio-selected survey. Using a radio-selected sample with broadband photometry at infrared and optical wavelengths, we test the k-Nearest Neighbours (kNN) and Random Forest machine learning algorithms, testing them both in their regression and classification modes. Further, we test different distance metrics used by the kNN algorithm, including the standard Euclidean distance, the Mahalanobis distance and a learned distance metric for both the regression mode (the Metric Learning for Kernel Regression metric) and the classification mode (the Large Margin Nearest Neighbour metric). We find that all regression-based modes fail on galaxies at a redshift z > 1. However, below this range, the kNN algorithm using the Mahalanobis distance metric performs best, with an η 0.15 outlier rate of 5.85%. In the classification mode, the kNN algorithm using the Mahalanobis distance metric also performs best, with an η 0.15 outlier rate of 5.85%, correctly placing 74% of galaxies in the top z > 1.02 bin. Finally, we also tested the effect of training in one field and applying the trained algorithm to similar data from another field and found that variation across fields does not result in statistically significant differences in predicted redshifts. Importantly, we find that while we may not be able to predict a continuous value for high-redshift radio sources, we can identify the majority of them using the classification modes of existing techniques.

show abstract

QSO photometric redshifts using machine learning and neural networks

Cited by 25 publications

References 52 publications

Intervening or associated? Machine learning classification of redshifted H I 21-cm absorption

Intervening or associated? Machine learning classification of redshifted H I 21-cm absorption

On the scarcity of redshifted OH and millimetre-band molecular absorption

Estimating Galaxy Redshift in Radio-Selected Datasets using Machine Learning

Contact Info

Product

Resources

About