CMU DeepLens: Deep Learning For Automatic Image-based Galaxy-Galaxy Strong Lens Finding

CMU DeepLens: Deep Learning For Automatic Image-based Galaxy-Galaxy Strong Lens Finding. Lanusse, F., Ma, Q., Li, N., Collett, T. E., Li, C., Ravanbakhsh, S., Mandelbaum, R., & Poczos, B. ArXiv e-prints, 1703:arXiv:1703.02642, March, 2017.

Paper abstract bibtex

Galaxy-scale strong gravitational lensing is not only a valuable probe of the dark matter distribution of massive galaxies, but can also provide valuable cosmological constraints, either by studying the population of strong lenses or by measuring time delays in lensed quasars. Due to the rarity of galaxy-scale strongly lensed systems, fast and reliable automated lens finding methods will be essential in the era of large surveys such as LSST, Euclid, and WFIRST. To tackle this challenge, we introduce CMU DeepLens, a new fully automated galaxy-galaxy lens finding method based on Deep Learning. This supervised machine learning approach does not require any tuning after the training step which only requires realistic image simulations of strongly lensed systems. We train and validate our model on a set of 20,000 LSST-like mock observations including a range of lensed systems of various sizes and signal-to-noise ratios (S/N). We find on our simulated data set that for a rejection rate of non-lenses of 99%, a completeness of 90% can be achieved for lenses with Einstein radii larger than 1.4" and S/N larger than 20 on individual \$g\$-band LSST exposures. Finally, we emphasize the importance of realistically complex simulations for training such machine learning methods by demonstrating that the performance of models of significantly different complexities cannot be distinguished on simpler simulations. We make our code publicly available at https://github.com/McWilliamsCenter/CMUDeepLens .

@article{lanusse_cmu_2017,
	title = {{CMU} {DeepLens}: {Deep} {Learning} {For} {Automatic} {Image}-based {Galaxy}-{Galaxy} {Strong} {Lens} {Finding}},
	volume = {1703},
	shorttitle = {{CMU} {DeepLens}},
	url = {http://adsabs.harvard.edu/abs/2017arXiv170302642L},
	abstract = {Galaxy-scale strong gravitational lensing is not only a valuable probe 
of the dark matter distribution of massive galaxies, but can also
provide valuable cosmological constraints, either by studying the
population of strong lenses or by measuring time delays in lensed
quasars. Due to the rarity of galaxy-scale strongly lensed systems, fast
and reliable automated lens finding methods will be essential in the era
of large surveys such as LSST, Euclid, and WFIRST. To tackle this
challenge, we introduce CMU DeepLens, a new fully automated
galaxy-galaxy lens finding method based on Deep Learning. This
supervised machine learning approach does not require any tuning after
the training step which only requires realistic image simulations of
strongly lensed systems. We train and validate our model on a set of
20,000 LSST-like mock observations including a range of lensed systems
of various sizes and signal-to-noise ratios (S/N). We find on our
simulated data set that for a rejection rate of non-lenses of 99\%, a
completeness of 90\% can be achieved for lenses with Einstein radii
larger than 1.4" and S/N larger than 20 on individual \$g\$-band LSST
exposures. Finally, we emphasize the importance of realistically complex
simulations for training such machine learning methods by demonstrating
that the performance of models of significantly different complexities
cannot be distinguished on simpler simulations. We make our code
publicly available at https://github.com/McWilliamsCenter/CMUDeepLens .},
	urldate = {2017-03-15},
	journal = {ArXiv e-prints},
	author = {Lanusse, Francois and Ma, Quanbin and Li, Nan and Collett, Thomas E. and Li, Chun-Liang and Ravanbakhsh, Siamak and Mandelbaum, Rachel and Poczos, Barnabas},
	month = mar,
	year = {2017},
	keywords = {Astrophysics - Astrophysics of Galaxies, Astrophysics - Cosmology and Nongalactic Astrophysics, Astrophysics - Instrumentation and Methods for Astrophysics},
	pages = {arXiv:1703.02642},
}

Downloads: 0

{"_id":"yLXHgRywwesuWYZDh","bibbaseid":"lanusse-ma-li-collett-li-ravanbakhsh-mandelbaum-poczos-cmudeeplensdeeplearningforautomaticimagebasedgalaxygalaxystronglensfinding-2017","author_short":["Lanusse, F.","Ma, Q.","Li, N.","Collett, T. E.","Li, C.","Ravanbakhsh, S.","Mandelbaum, R.","Poczos, B."],"bibdata":{"bibtype":"article","type":"article","title":"CMU DeepLens: Deep Learning For Automatic Image-based Galaxy-Galaxy Strong Lens Finding","volume":"1703","shorttitle":"CMU DeepLens","url":"http://adsabs.harvard.edu/abs/2017arXiv170302642L","abstract":"Galaxy-scale strong gravitational lensing is not only a valuable probe of the dark matter distribution of massive galaxies, but can also provide valuable cosmological constraints, either by studying the population of strong lenses or by measuring time delays in lensed quasars. Due to the rarity of galaxy-scale strongly lensed systems, fast and reliable automated lens finding methods will be essential in the era of large surveys such as LSST, Euclid, and WFIRST. To tackle this challenge, we introduce CMU DeepLens, a new fully automated galaxy-galaxy lens finding method based on Deep Learning. This supervised machine learning approach does not require any tuning after the training step which only requires realistic image simulations of strongly lensed systems. We train and validate our model on a set of 20,000 LSST-like mock observations including a range of lensed systems of various sizes and signal-to-noise ratios (S/N). We find on our simulated data set that for a rejection rate of non-lenses of 99%, a completeness of 90% can be achieved for lenses with Einstein radii larger than 1.4\" and S/N larger than 20 on individual \\$g\\$-band LSST exposures. Finally, we emphasize the importance of realistically complex simulations for training such machine learning methods by demonstrating that the performance of models of significantly different complexities cannot be distinguished on simpler simulations. We make our code publicly available at https://github.com/McWilliamsCenter/CMUDeepLens .","urldate":"2017-03-15","journal":"ArXiv e-prints","author":[{"propositions":[],"lastnames":["Lanusse"],"firstnames":["Francois"],"suffixes":[]},{"propositions":[],"lastnames":["Ma"],"firstnames":["Quanbin"],"suffixes":[]},{"propositions":[],"lastnames":["Li"],"firstnames":["Nan"],"suffixes":[]},{"propositions":[],"lastnames":["Collett"],"firstnames":["Thomas","E."],"suffixes":[]},{"propositions":[],"lastnames":["Li"],"firstnames":["Chun-Liang"],"suffixes":[]},{"propositions":[],"lastnames":["Ravanbakhsh"],"firstnames":["Siamak"],"suffixes":[]},{"propositions":[],"lastnames":["Mandelbaum"],"firstnames":["Rachel"],"suffixes":[]},{"propositions":[],"lastnames":["Poczos"],"firstnames":["Barnabas"],"suffixes":[]}],"month":"March","year":"2017","keywords":"Astrophysics - Astrophysics of Galaxies, Astrophysics - Cosmology and Nongalactic Astrophysics, Astrophysics - Instrumentation and Methods for Astrophysics","pages":"arXiv:1703.02642","bibtex":"@article{lanusse_cmu_2017,\n\ttitle = {{CMU} {DeepLens}: {Deep} {Learning} {For} {Automatic} {Image}-based {Galaxy}-{Galaxy} {Strong} {Lens} {Finding}},\n\tvolume = {1703},\n\tshorttitle = {{CMU} {DeepLens}},\n\turl = {http://adsabs.harvard.edu/abs/2017arXiv170302642L},\n\tabstract = {Galaxy-scale strong gravitational lensing is not only a valuable probe \nof the dark matter distribution of massive galaxies, but can also\nprovide valuable cosmological constraints, either by studying the\npopulation of strong lenses or by measuring time delays in lensed\nquasars. Due to the rarity of galaxy-scale strongly lensed systems, fast\nand reliable automated lens finding methods will be essential in the era\nof large surveys such as LSST, Euclid, and WFIRST. To tackle this\nchallenge, we introduce CMU DeepLens, a new fully automated\ngalaxy-galaxy lens finding method based on Deep Learning. This\nsupervised machine learning approach does not require any tuning after\nthe training step which only requires realistic image simulations of\nstrongly lensed systems. We train and validate our model on a set of\n20,000 LSST-like mock observations including a range of lensed systems\nof various sizes and signal-to-noise ratios (S/N). We find on our\nsimulated data set that for a rejection rate of non-lenses of 99\\%, a\ncompleteness of 90\\% can be achieved for lenses with Einstein radii\nlarger than 1.4\" and S/N larger than 20 on individual \\$g\\$-band LSST\nexposures. Finally, we emphasize the importance of realistically complex\nsimulations for training such machine learning methods by demonstrating\nthat the performance of models of significantly different complexities\ncannot be distinguished on simpler simulations. We make our code\npublicly available at https://github.com/McWilliamsCenter/CMUDeepLens .},\n\turldate = {2017-03-15},\n\tjournal = {ArXiv e-prints},\n\tauthor = {Lanusse, Francois and Ma, Quanbin and Li, Nan and Collett, Thomas E. and Li, Chun-Liang and Ravanbakhsh, Siamak and Mandelbaum, Rachel and Poczos, Barnabas},\n\tmonth = mar,\n\tyear = {2017},\n\tkeywords = {Astrophysics - Astrophysics of Galaxies, Astrophysics - Cosmology and Nongalactic Astrophysics, Astrophysics - Instrumentation and Methods for Astrophysics},\n\tpages = {arXiv:1703.02642},\n}\n\n","author_short":["Lanusse, F.","Ma, Q.","Li, N.","Collett, T. E.","Li, C.","Ravanbakhsh, S.","Mandelbaum, R.","Poczos, B."],"key":"lanusse_cmu_2017","id":"lanusse_cmu_2017","bibbaseid":"lanusse-ma-li-collett-li-ravanbakhsh-mandelbaum-poczos-cmudeeplensdeeplearningforautomaticimagebasedgalaxygalaxystronglensfinding-2017","role":"author","urls":{"Paper":"http://adsabs.harvard.edu/abs/2017arXiv170302642L"},"keyword":["Astrophysics - Astrophysics of Galaxies","Astrophysics - Cosmology and Nongalactic Astrophysics","Astrophysics - Instrumentation and Methods for Astrophysics"],"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://bibbase.org/zotero/polyphant","dataSources":["7gvjSdWrEu7z5vjjj"],"keywords":["astrophysics - astrophysics of galaxies","astrophysics - cosmology and nongalactic astrophysics","astrophysics - instrumentation and methods for astrophysics"],"search_terms":["cmu","deeplens","deep","learning","automatic","image","based","galaxy","galaxy","strong","lens","finding","lanusse","ma","li","collett","li","ravanbakhsh","mandelbaum","poczos"],"title":"CMU DeepLens: Deep Learning For Automatic Image-based Galaxy-Galaxy Strong Lens Finding","year":2017}