Artificial Intelligence And Machine Learning In Astronomy

3y ago
3.60 MB
72 Pages
Last View : 18d ago
Last Download : 4m ago
Upload by : Jayda Dunning

Artificial Intelligence andMachine Learning in AstronomyOfer Lahav (UCL)1

Introduction card2

Artificial Neural Networks:early days3

Life 3.0Tegmark’s book (2018)4

What is ‘Big Data’? Wikipedia’s definition: “data sets that are so large orcomplex that TRADITIONAL data processingapplications are inadequate to deal with them”. Clearly, this is a ‘moving target’. "Big data is high volume, high velocity, and/or highvariety information assets that require new forms ofprocessing to enable enhanced decision making,insight discovery and process optimization.”(Gartner)5

Can we trust just the humanbrain?(can you see 12 black dots at once?)6

Machine Learning7

Machine Learning MethodsDecision TreesArtificial Neural Networks8

On a Deep Learning Curve credit:Y. LeCuncredit: A. Ng9

Artificial Intelligence, Machine Learning, DeepLearning: are they ‘explainable’ ?10

Astro papers on the arXiv with Deep Learning’ in titleYear #Papers 2019: 83 2018: 35 2017: 2311

Big Data in AstronomySurveyData volume pernight/dayGalaxiesCostScientistsDES (2012-)1 TeraB 300 Million 40M 400DESI (2019-)40 GigaB 35 Million 70M 600LSST (2021-)15 TeraB 1 Billion 1.0B 1000Euclid (2021-)850 GigaB 1 Billion 1.5B 15001 PetaB 1 Billion 1.3B 1000SKA (2020-)12

Galaxy surveys timelineLSSTDESI1 of 6lenses@ UCLDESI had its first light in October 2019Euclid13

SKA Big Data Challengewww.skatelescope.org14

Machine Learning in Astronomy Machine learning examples from Astronomy:- Classification:galaxy type, star/galaxy, Supernovae Ia,strong gravitational lensing- Photo-z- Mass of the Local Group- The search for Planet 9 and exo-planets- Gravitational Waves & follow-ups- Likelihood-free parameter estimationDeep Learning15

What accelerates the Universe?“a simple but strange universe”16

Einstein 1917 LambdaModified NewtonianModified GRIn a static universe:Einstein (February 1917)English translation: ns/433?ajax17

The 1919 EclipseEddington’s experiment

Standard candles:Supernovae Ia

Probes of Dark EnergyStandard candlesStandard rulersGravitational Lensing1.0%45204080160DES SV mass map(Chang et al. 2016)0.5%500.0%55-0.5%matter density E [compared to cosmic mean]Clusterscluster richness20606:00h6:30h5:00h5:30h4:00h-1.0%

The Bayesian approach21Cf. Planck results 2018Credit: Jason McEwen

Open Questions on Dark EnergyDE equation of state:Pressure/density w(a) w0 wa (1-a)Is there a fundamental reason for w -1 (Lambda)?Is it on the LHS or RHS of Einstein’s equation?Is there a physical case for w -1?What is the case for a time-dependent w(z) ?When should we stop with w?(note ‘precision’ vs ‘accuracy’, cf. curvature)w Does Anthropic reasoning make sense?w Is a higher level theory to be discovered,connecting GR to Quantum Mechanics andThermodynamics? Will it take another 100 years ?wwwww22

The Dark Energy Survey**Multi-probe approachWide field: Cluster Counts,Weak Lensing, Large Scale StructureTime domain: SupernovaeCTIOSurvey strategy- 300 million galaxies with photometric redshifts- 2500 SN Ia*Over 400 scientists based in 7 countries*6 seasons of observations completed 758 nights in total* Over 250 DES papers on the arXiv* DES book23

The DES bookThis book is about the Dark Energy Survey, a cosmological experimentdesigned to investigate the physical nature of dark energy bymeasuring its effect on the expansion history of the universe and onthe growth of large-scale structure. The survey saw first light in 2012,after a decade of planning, and completed observations in 2019.The collaboration designed and built a 570-megapixel camera andinstalled it on the four-metre Blanco telescope at the Cerro TololoInter-American Observatory in the Chilean Andes. The survey datayielded a three-dimensional map of over 300 million galaxies and acatalogue of thousands of supernovae. Analysis of the early data hasconfirmed remarkably accurately the model of cold dark matter anda cosmological constant. The survey has also offered new insightsinto galaxies, supernovae, stellar evolution, solar system objects andthe nature of gravitational wave events.A project of this scale required the long-term commitment of hundredsof scientists from institutions all over the world. The chapters in thefirst three sections of the book were either written by these scientistsor based on interviews with them. These chapters explain, for a nonspecialist reader, the science analysis involved. They also describehow the project was conceived, and chronicle some of the many anddiverse challenges involved in advancing our understanding of theuniverse. The final section is trans-disciplinary, including inputs froma philosopher, an anthropologist, visual artists and a poet. Scientificcollaborations are human endeavours and the book aims to conveya sense of the wider context within which science comes about.This book is addressed to scientists, decision makers, socialscientists and engineers, as well as to anyone with an interest incontemporary cosmology and astrophysics.The Dark Energy SurveyThe Dark Energy SurveyThe Story of a Cosmological ExperimentTheDark EnergySurveyThe Story of aCosmological ExperimentOfer LahavLucy CalderJulian MayersJosh FriemanEditorsLahavCalderMayersFriemanCover photo: Reidar Hahn, Fermilab.World Scientificwww.worldscientific.comQ0247 hcISBN 978-1-78634-835-7World Scientific24

207 DES SN Ia( 122 other SN Ia)DES collaboration, 1811.02374w 0.9780.059, and Ω 0.321m0.018 (1-sigma)Blinding to overcome confirmation bias25

from DES Planck BAO SNIaw -1.00 -0.05 0.04Neutrino mass 0.29 eVnote 20 nuisance parameters:Clumpiness amplitude3x2pt statistic: DES Year 1 (1300 sq deg) resultsfrom galaxy clustering (650K LRGs)and weak lensing (26M source galaxies)matter density26arXiv:1708.01530 (and follow up multi-probe; extensions)

H0 TensionarXiv:1907.1062527

Will LCDM survive?(I) Will the tension in(a) H0 (ladder vs CMB) 4 sigma(b) S8-Omegam (WL vs CMB) 2 sigmago away after more ‘bread and butter’ work?(II) If the tension remains/grows, would it leadto new Physics or a departure from LCDM?28

The first Black Hole Binarydetected by LIGOGW15091429

Gravitational Waves:The visible light fromthe Kilonova fading awayGalaxy NGC 4993, 40Mpc away30

The Hubble constant H0 fromGW170817w Hubble Constant from GW standard siren:H0 vH/d 70 ( 12 -8) km/sec/MpcWith these 68% CL, consistent with bothPlanck and SNIa, which are in tension with eachother.31Abbott et al, Nature 2017

The Impact of Peculiar Velocities on H0from Gravitational Wave Bright SirensConstantina Nicolaou, OL, et al. 1909.09609GW 170817 in NGC4993At distance of 40 Mpc,Uncertainty of 200 km/seccorresponds to 4km/sec/MpcH0 68.6 14.0-8.5 km/sec/MpcBayesian Marginalizationover smoothing scales Cf. Abbott et al (2019)Howlett & Davis (2019),Mukherjee et al. (2019), 32

H0 from one Dark Siren 77k DES galaxies33Soaers-Santos, Palmese, Hartley.OL & DES, LVC; 1901.01540

(i) Object Classification with ML34

Exo-planet space missionsThe Futureis bright!

Machine Learning for detecting Exo-planetsConvolutionalNeural NetsClassificationPlanetNo PlanetYip, Waldmann et al. (2019)

Star/galaxy separation in DESSoumagnac et al (1306.5236)37

One Million galaxies classified by 100,000people!Lintott et al.38

Galaxy zoo and machine learningBanerji, OL et al. (0908.2033)Cf. OL, Naim et al. (1995)39

Photometric redshift Probe strongspectral features(4000 break) Difference in fluxthrough filters as thegalaxy is redshifted.40

Photo-Z codesCODEMETHODREFERENCEHyperZTemplateBolzonella et al. (2000)BPZTPZBayesianTreesBenitez (2000)Carraso Kind & Brunner(2013)ANNz1ANNz2TrainingCollister & Lahav (2004)Sadeh, Abdalla & Lahav(2016)ZEBRAHybrid, BayesianFeldmann et al. (2006)LePhareTemplateIlbert et al. (2006)41

Photo-z: DES SV dataSanchez et al. (2015)Bonnett et al. (2015)incl. new ANNz2,Sadeh, Abdalla & OL (2016)42

End-to-end: the impact of differentPhZ codes on DES-SV WLDES collaboration 201543

Finding Strong Lensing Arcswith Machine Learning HST image of cluster Data ChallengeSDSS J1038 4849Metcalf et al.c ESO 2018Astronomy & Astrophysics manuscript no. paperFebruary 20, 2018The Strong Gravitational Lens Finding ChallengearXiv:1802.03609v2 [astro-ph.GA] 17 Feb 2018R. Benton Metcalf1, 2? , M. Meneghetti2 , Camille Avestruz3, 4, 5,? , Fabio Bellagamba1, 2 , Clécio R. Bom6, 7 ,Emmanuel Bertin8 , Rémi Cabanac9 , Andrew Davies22 , Etienne Decencière10 , Rémi Flamary11 , RaphaelGavazzi8 , Mario Geiger12 , Philippa Hartley13 , Marc Huertas-Company14 , Neal Jackson13 , Eric Jullo15 ,Jean-Paul Kneib12 , Léon V. E. Koopmans16 , François Lanusse17 , Chun-Liang Li18 , Quanbin Ma18 , MartinMakler7 , Nan Li19 , Matthew Lightman15 , Carlo Enrico Petrillo16 , Stephen Serjeant22 , Christoph Schäfer12 ,Alessandro Sonnenfeld21 , Amit Tagore13 , Crescenzo Tortora16 , Diego Tuccillo10, 14 , Manuel B. Valentı́n7 ,Santiago Velasco-Forero10 , Gijs A. Verdoes Kleijn16 , and Georgios rtimento di Fisica & Astronomia, Università di Bologna, via Gobetti 93/2, 40129 Bologna, ItalyINAF-Osservatorio Astronomico di Bologna, via Ranzani 1, 40127 Bologna, ItalyEnrico Fermi Institute, The University of Chicago, Chicago, IL 60637 U.S.A.Kavli Institute for Cosmological Physics, The University of Chicago, Chicago, IL 60637 U.S.A.Department of Astronomy & Astrophysics, The University of Chicago, Chicago, IL 60637 U.S.A.Centro Federal de Educação Tecnológica Celso Suckow da Fonseca, CEP 23810-000, Itaguaı́, RJ, BrazilCentro Brasileiro de Pesquisas Fı́sicas, CEP 22290-180, Rio de Janeiro, RJ, BrazilInstitut d’Astrophysique de Paris, Sorbonne Université, CNRS, UMR 7095, 98 bis bd Arago, 75014 Paris, France.IRAP, Université de Toulouse, CNRS, UPS, Toulouse, France.MINES Paristech, PSL Research University, Centre for Mathematical Morphology, 35 rue Saint-Honore,Fontainebleau, FranceLaboratoire Lagrange, Universié de Nice Sophia-Antipolis, Centre National de la Recherche Scientifique,Institute of Physics, Laboratory of Astrophysics, Ecole Polytechnique Fédérale de Lausanne (EPFL), Observatoirede Sauverny, 1290 Versoix, SwitzerlandJodrell Bank Centre for Astrophysics, School of Physics & Astronomy, University of Manchester, Oxford Rd, Manchester M13 9PL, UK Observatoire de la Côte d’Azur, Parc Valrose, 06108 Nice, FranceLERMA, Observatoire de Paris, CNRS, Université Paris Diderot, 61, Avenue de l’Observatoire F-75014, Paris,FranceAix Marseille Université, CNRS, LAM (Laboratoire d’Astrophysique de Marseille) UMR 7326, 13388, Marseille,FranceKapteyn Astronomical Institute, University of Groningen, Postbus 800, 9700 AV, Groningen, The NetherlandsMcWilliams Center for Cosmology, Department of Physics, Carnegie Mellon University, Pittsburgh, PA 15213, USASchool of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USASchool of Physics and Astronomy, Nottingham University, University Park, Nottingham, NG7 2RD, UKJPMorgan Chase, Chicago, IL 60603 U.S.A.Kavli IPMU (WPI), UTIAS, The University of Tokyo, Kashiwa, Chiba 277-8583, JapanSchool of Physical Sciences, The Open University, Walton Hall, Milton Keynes, MK7 6AA, UKFebruary 20, 2018ABSTRACTLarge scale imaging surveys will increase the number of galaxy-scale strong lensing candidates by maybe three ordersof magnitudes beyond the number known today. Finding these rare objects will require picking them out of at leasttens of millions of images and deriving scientific results from them will require quantifying the efficiency and bias ofany search method. To achieve these objectives automated methods must be developed. Because gravitational lensesare rare objects reducing false positives will be particularly important. We present a description and results of an opengravitational lens finding challenge. Participants were asked to classify 100,000 candidate objects as to whether theywere gravitational lenses or not with the goal of developing better automated methods for finding lenses in large datasets. A variety of methods were used including visual inspection, arc and ring finders, support vector machines (SVM)and convolutional neural networks (CNN). We find that many of the methods will be easily fast enough to analyse theanticipated data flow. In test data, several methods are able to identify upwards of half the lenses after applying somethresholds on the lens characteristics such as lensed image brightness, size or contrast with the lens galaxy withoutmaking a single false-positive identification. This is significantly better than direct inspection by humans was able todo. Having multi-band, ground based data is found to be better for this purpose than single-band space based datawith lower noise and higher resolution, suggesting that colours bring a crucial additional information. The most difficultchallenge for a lens finder is di erentiating between rare irregular and ring-like face-on galaxies and true gravitationallenses. The degree to which the efficiency and biases of lens finders can be quantified largely depends on the realism ofthe simulated data on which the finders are trained.Article number, page 1 of 2244

CMUDeepLens (Lanusse et al. 1703.02642) Mocks with arcs Mocks without arcsExpected in LSST: about one million strongly lensed galaxies out of an estimated 20billion galaxies.45The approach: supervised CNN. Completeness of 90% can be achieved

(ii) Time Domain with ML46

Light-curve feature selection47

Photometric Classification ofSupernovaeLochner, McEwen,Peiris, Lahav, Winter48arXiv: 1603.00882

Feature extraction with Wavelet 6 classifiersLochner et al. (2016)49

(iii) Map reconstruction50

Mass mapping from DES WLSparsity prior (Starck et al. 2015)N. Jeffery et al.arXiv:1801.08

5 Department of Astronomy & Astrophysics, The University of Chicago, Chicago, IL 60637 U.S.A. 6 Centro Federal de Educac a o Tecnolo gica Celso Suckow da Fonseca, CEP 23810-000, Itagua ı, RJ, Brazil 7 Centro Brasileiro de Pesquisas F ısicas, CEP 22290-180, Rio de Janeiro, RJ, Brazil 8 Institut d’Astrophysique de Paris, Sorbonne Universit e, CNRS, UMR 7095, 98 bis bd Arago, 75014 .

Related Documents:

Artificial Intelligence, Machine Learning, and Deep Learning (AI/ML/DL) F(x) Deep Learning Artificial Intelligence Machine Learning Artificial Intelligence Technique where computer can mimic human behavior Machine Learning Subset of AI techniques which use algorithms to enable machines to learn from data Deep Learning

Artificial Intelligence -a brief introduction Project Management and Artificial Intelligence -Beyond human imagination! November 2018 7 Artificial Intelligence Applications Artificial Intelligence is the ability of a system to perform tasks through intelligent deduction, when provided with an abstract set of information.

This Artificial Intelligence Master's Program provides training for the most sought after skills required for a successful career in Artificial Intelligence. As you undertake your Artificial Intelligence Engineer training, you will master the concepts of Deep Learning, Machine Learning, Natural Language Processing plus

This Artificial Intelligence Master's Program covers the crucial skills you need for a successful career in artificial intelligence (AI). As you undertake your Artificial Intelligence course, you'll master the concepts of the machine and deep learning—plus the programming languages needed to excel in an AI career with exclusive

CS8691-Artificial Intelligence. 23 of 32 Learning to Imitate a Human Action CS8691-Artificial Intelligence. 24 of 32 Result after Learning HumanAction Imitation CS8691-Artificial Intelligence. 25 of 32 The Future of AI Massiv

BCS Foundation Certificate in Artificial Intelligence V1.1 Oct 2020 Syllabus Learning Objectives 1. Ethical and Sustainable Human and Artificial Intelligence (20%) Candidates will be able to: 1.1. Recall the general definition of Human and Artificial Intelligence (AI). 1.1.1. Describe the concept of intelligent agents. 1.1.2. Describe a modern .

IN ARTIFICIAL INTELLIGENCE Stuart Russell and Peter Norvig, Editors FORSYTH & PONCE Computer Vision: A Modern Approach GRAHAM ANSI Common Lisp JURAFSKY & MARTIN Speech and Language Processing, 2nd ed. NEAPOLITAN Learning Bayesian Networks RUSSELL & NORVIG Artificial Intelligence: A Modern Approach, 3rd ed. Artificial Intelligence A Modern Approach Third Edition Stuart J. Russell and Peter .

and artificial intelligence expert, joined Ernst & Young as the person in charge of its global innovative artificial intelligence team. In recent years, many countries have been competing to carry out research and application of artificial intelli-gence, and the call for he use of artificial