Classifying the formation processes of s0 galaxies using convolutional neural networks

J. D. Diaz, Kenji Bekki, Duncan A. Forbes, Warrick J. Couch, Michael J. Drinkwater, Simon Deeley

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Numerous studies have demonstrated the ability of Convolutional Neural Networks (CNNs) to classify large numbers of galaxies in a manner that mimics the expertise of astronomers. Such classifications are not always physically motivated, however, such as categorizing galaxies by their morphological types. In this work, we consider the use of CNNs to classify simulated S0 galaxies based on fundamental physical properties. In particular, we undertake two investigations: (1) the classification of simulated S0 galaxies into three distinct evolutionary paths (isolated, tidal interaction in a group halo, and spiral-spiralmerger) and (2) the prediction of the mass ratio for the S0s formed via mergers. To train the CNNs, we first run several hundred N-body simulations to model the formation of S0s under idealized conditions, and then we build our training data sets by creating images of stellar density and two-dimensional kinematic maps for each simulated S0. Our trained networks have remarkable accuracies exceeding 99 per cent when classifying the S0 formation pathway. For the case of predicting merger mass ratios, the mean predictions are consistent with the true values to within roughly one standard deviation across the full range of our data. Our work demonstrates the potential of CNNs to classify galaxies by the fundamental physical properties that drive their evolution.

Original languageEnglish
Pages (from-to)4845-4862
Number of pages18
JournalMonthly Notices of the Royal Astronomical Society
Volume486
Issue number4
DOIs
Publication statusPublished - 1 Jul 2019

Fingerprint

classifying
merger
physical property
galaxies
prediction
train
kinematics
mass ratios
physical properties
simulation
predictions
standard deviation
halos
education
interactions

Cite this

Diaz, J. D. ; Bekki, Kenji ; Forbes, Duncan A. ; Couch, Warrick J. ; Drinkwater, Michael J. ; Deeley, Simon. / Classifying the formation processes of s0 galaxies using convolutional neural networks. In: Monthly Notices of the Royal Astronomical Society. 2019 ; Vol. 486, No. 4. pp. 4845-4862.
@article{712549dfadb54bceb1bff8f9c7bdaf5e,
title = "Classifying the formation processes of s0 galaxies using convolutional neural networks",
abstract = "Numerous studies have demonstrated the ability of Convolutional Neural Networks (CNNs) to classify large numbers of galaxies in a manner that mimics the expertise of astronomers. Such classifications are not always physically motivated, however, such as categorizing galaxies by their morphological types. In this work, we consider the use of CNNs to classify simulated S0 galaxies based on fundamental physical properties. In particular, we undertake two investigations: (1) the classification of simulated S0 galaxies into three distinct evolutionary paths (isolated, tidal interaction in a group halo, and spiral-spiralmerger) and (2) the prediction of the mass ratio for the S0s formed via mergers. To train the CNNs, we first run several hundred N-body simulations to model the formation of S0s under idealized conditions, and then we build our training data sets by creating images of stellar density and two-dimensional kinematic maps for each simulated S0. Our trained networks have remarkable accuracies exceeding 99 per cent when classifying the S0 formation pathway. For the case of predicting merger mass ratios, the mean predictions are consistent with the true values to within roughly one standard deviation across the full range of our data. Our work demonstrates the potential of CNNs to classify galaxies by the fundamental physical properties that drive their evolution.",
keywords = "Galaxies: elliptical and lenticular, Galaxies: kinematics and dynamics",
author = "Diaz, {J. D.} and Kenji Bekki and Forbes, {Duncan A.} and Couch, {Warrick J.} and Drinkwater, {Michael J.} and Simon Deeley",
year = "2019",
month = "7",
day = "1",
doi = "10.1093/mnras/stz1057",
language = "English",
volume = "486",
pages = "4845--4862",
journal = "Monthly Notices of the Royal Astronomical Society",
issn = "0035-8711",
publisher = "OXFORD UNIV PRESS UNITED KINGDOM",
number = "4",

}

Classifying the formation processes of s0 galaxies using convolutional neural networks. / Diaz, J. D.; Bekki, Kenji; Forbes, Duncan A.; Couch, Warrick J.; Drinkwater, Michael J.; Deeley, Simon.

In: Monthly Notices of the Royal Astronomical Society, Vol. 486, No. 4, 01.07.2019, p. 4845-4862.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Classifying the formation processes of s0 galaxies using convolutional neural networks

AU - Diaz, J. D.

AU - Bekki, Kenji

AU - Forbes, Duncan A.

AU - Couch, Warrick J.

AU - Drinkwater, Michael J.

AU - Deeley, Simon

PY - 2019/7/1

Y1 - 2019/7/1

N2 - Numerous studies have demonstrated the ability of Convolutional Neural Networks (CNNs) to classify large numbers of galaxies in a manner that mimics the expertise of astronomers. Such classifications are not always physically motivated, however, such as categorizing galaxies by their morphological types. In this work, we consider the use of CNNs to classify simulated S0 galaxies based on fundamental physical properties. In particular, we undertake two investigations: (1) the classification of simulated S0 galaxies into three distinct evolutionary paths (isolated, tidal interaction in a group halo, and spiral-spiralmerger) and (2) the prediction of the mass ratio for the S0s formed via mergers. To train the CNNs, we first run several hundred N-body simulations to model the formation of S0s under idealized conditions, and then we build our training data sets by creating images of stellar density and two-dimensional kinematic maps for each simulated S0. Our trained networks have remarkable accuracies exceeding 99 per cent when classifying the S0 formation pathway. For the case of predicting merger mass ratios, the mean predictions are consistent with the true values to within roughly one standard deviation across the full range of our data. Our work demonstrates the potential of CNNs to classify galaxies by the fundamental physical properties that drive their evolution.

AB - Numerous studies have demonstrated the ability of Convolutional Neural Networks (CNNs) to classify large numbers of galaxies in a manner that mimics the expertise of astronomers. Such classifications are not always physically motivated, however, such as categorizing galaxies by their morphological types. In this work, we consider the use of CNNs to classify simulated S0 galaxies based on fundamental physical properties. In particular, we undertake two investigations: (1) the classification of simulated S0 galaxies into three distinct evolutionary paths (isolated, tidal interaction in a group halo, and spiral-spiralmerger) and (2) the prediction of the mass ratio for the S0s formed via mergers. To train the CNNs, we first run several hundred N-body simulations to model the formation of S0s under idealized conditions, and then we build our training data sets by creating images of stellar density and two-dimensional kinematic maps for each simulated S0. Our trained networks have remarkable accuracies exceeding 99 per cent when classifying the S0 formation pathway. For the case of predicting merger mass ratios, the mean predictions are consistent with the true values to within roughly one standard deviation across the full range of our data. Our work demonstrates the potential of CNNs to classify galaxies by the fundamental physical properties that drive their evolution.

KW - Galaxies: elliptical and lenticular

KW - Galaxies: kinematics and dynamics

UR - http://www.scopus.com/inward/record.url?scp=85067933359&partnerID=8YFLogxK

U2 - 10.1093/mnras/stz1057

DO - 10.1093/mnras/stz1057

M3 - Article

VL - 486

SP - 4845

EP - 4862

JO - Monthly Notices of the Royal Astronomical Society

JF - Monthly Notices of the Royal Astronomical Society

SN - 0035-8711

IS - 4

ER -