Separating stars from quasars: Machine learning investigation using photometric data

Makhija, Simran; Saha, Snehanshu; Basak, Suryoday; Das, Mousumi

doi:10.1016/j.ascom.2019.100313

Cited by 17 publications

(12 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…the boundary definition, or which photometric bands and aperture types to use -are often based on a few low-dimensional projections of a complex high-dimensional space. On the other hand, modern machine learning methods enable us to automatically make such choices in a data-driven way, utilizing the entire multidimensional parameter space (Gao, Zhang & Zhao 2008;Kim, Brunner & Carrasco Kind 2015;Makhija et al 2019).…”

Section: Introductionmentioning

confidence: 99%

PS1-STRM: Neural network source classification and photometric redshift catalogue for PS1 $3π$ DR1

Beck,

Szapudi,

Flewelling

et al. 2019

Preprint

View full text Add to dashboard Cite

The Pan-STARRS1 (PS1) 3π survey is a comprehensive optical imaging survey of three quarters of the sky in the grizy broad-band photometric filters. We present the methodology used in assembling the source classification and photometric redshift (photo-z ) catalogue for PS1 3π Data Release 1, titled Pan-STARRS1 Source Types and Redshifts with Machine learning (PS1-STRM).For both main data products, we use neural network architectures, trained on a compilation of public spectroscopic measurements that has been cross-matched with PS1 sources.We quantify the parameter space coverage of our training data set, and flag extrapolation using self-organizing maps. We perform a Monte-Carlo sampling of the photometry to estimate photo-z uncertainty.The final catalogue contains 2, 902, 054, 648 objects. On our validation data set, for non-extrapolated sources, we achieve an overall classification accuracy of 98.1% for galaxies, 97.8% for stars, and 96.6% for quasars.Regarding the galaxy photo-z estimation, we attain an overall bias of ∆z norm = 0.0005, a standard deviation of σ(∆z norm ) = 0.0322, a median absolute deviation of MAD(∆z norm ) = 0.0161, and an outlier fraction of O = 1.89%.The catalogue will be made available as a high-level science product via the Mikulski Archive for Space Telescopes at https://doi.org/10.17909//t9-rnk7-gr88.

show abstract

Section: Introductionmentioning

confidence: 99%

PS1-STRM: Neural network source classification and photometric redshift catalogue for PS1 $3π$ DR1

Beck,

Szapudi,

Flewelling

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…As further proof of the efficacy of our proposed activation function, prior work by Saha and his group, Makhija et al [56] uses a special case of our activation function in the discriminator network of a Generative Adversarial Network (GAN) to achieve exemplary performance in separating stars from quasars, outperforming the same architecture with the sigmoid activation. We defer to Makhija et al [56] for a more detailed explanation of the method. The same activation was also used on the Gene Expression Omnibus (GEO) dataset by Sridhar et al [57], and was found to outperform both the sigmoid and ReLU activations.…”

Section: Monotonicity Of Sbafmentioning

confidence: 87%

Evolution of novel activation functions in neural network training for astronomy data: habitability classification of exoplanets

Saha

Nagaraj

Mathur³

et al. 2020

Eur. Phys. J. Spec. Top.

View full text Add to dashboard Cite

“…With the advent of large surveys like the Sloan Digital Sky Survey (SDSS, York et al 2000), optical data could be coupled with space-based MIR data to find the stellar locus in a 10-dimensional color-space (Davenport et al 2014). Recent efforts to separate stars from quasars, or perform a regression on effective temperature with machine learning on photometric data have been successful (Makhija et al 2019;Bai et al 2019); however, these studies are often focused on main sequence, low mass stars. This is an understandable choice given the rarity of evolved, high mass stars, the absence of reliable distances to calculate luminosities from which to select putative massive stars, and the fact that followup spectroscopy is necessary in order to confirm a star's membership in many important classes.…”

Section: Classifier Selectionmentioning

confidence: 99%

Photometric Classifications of Evolved Massive Stars: Preparing for the Era of Webb and Roman with Machine Learning

Dorn-Wallenstein,

Davenport,

Huppenkothen

et al. 2021

Preprint

View full text Add to dashboard Cite

In the coming years, next-generation space-based infrared observatories will significantly increase our samples of rare massive stars, representing a tremendous opportunity to leverage modern statistical tools and methods to test massive stellar evolution in entirely new environments. Such work is only possible if the observed objects can be reliably classified. Spectroscopic observations are infeasible with more distant targets, and so we wish to determine whether machine learning methods can classify massive stars using broadband infrared photometry. We find that a Support Vector Machine classifier is capable of coarsely classifying massive stars with labels corresponding to hot, cool, and emission line stars with high accuracy, while rejecting contaminating low mass giants. Remarkably, 76% of emission line stars can be recovered without the need for narrowband or spectroscopic observations. We classify a sample of ∼2500 objects with no existing labels, and identify fourteen candidate emission line objects. Unfortunately, despite the high precision of the photometry in our sample, the heterogeneous origins of the labels for the stars in our sample severely inhibits our classifier from distinguishing classes of stars with more granularity. Ultimately, no large and homogeneously labeled sample of massive stars currently exists. Without significant efforts robustly classify evolved massive stars -which is feasible given existing data from large all-sky spectroscopic surveys -shortcomings in the labeling of existing data sets will hinder efforts to leverage the next-generation of space observatories.

show abstract

Separating stars from quasars: Machine learning investigation using photometric data

Cited by 17 publications

References 24 publications

PS1-STRM: Neural network source classification and photometric redshift catalogue for PS1 $3π$ DR1

PS1-STRM: Neural network source classification and photometric redshift catalogue for PS1 $3π$ DR1

Evolution of novel activation functions in neural network training for astronomy data: habitability classification of exoplanets

Photometric Classifications of Evolved Massive Stars: Preparing for the Era of Webb and Roman with Machine Learning

Contact Info

Product

Resources

About