Encoding large scale cosmological structure with Generative Adversarial Networks

Encoding large scale cosmological structure with Generative Adversarial Networks. Ullmo, M., Decelle, A., & Aghanim, N. arXiv e-prints, 2011:arXiv:2011.05244, November, 2020.

Paper abstract bibtex

Recently a type of neural networks called Generative Adversarial Networks (GANs) has been proposed as a solution for fast generation of simulation-like datasets, in an attempt to bypass heavy computations and expensive cosmological simulations to run in terms of time and computing power. In the present work, we build and train a GAN to look further into the strengths and limitations of such an approach. We then propose a novel method in which we make use of a trained GAN to construct a simple autoencoder (AE) as a first step towards building a predictive model. Both the GAN and AE are trained on images issued from two types of N-body simulations, namely 2D and 3D simulations. We find that the GAN successfully generates new images that are statistically consistent with the images it was trained on. We then show that the AE manages to efficiently extract information from simulation images, satisfyingly inferring the latent encoding of the GAN to generate an image with similar large scale structures.

@article{ullmo_encoding_2020,
	title = {Encoding large scale cosmological structure with {Generative} {Adversarial} {Networks}},
	volume = {2011},
	url = {http://adsabs.harvard.edu/abs/2020arXiv201105244U},
	abstract = {Recently a type of neural networks called Generative Adversarial 
Networks (GANs) has been proposed as a solution for fast generation of
simulation-like datasets, in an attempt to bypass heavy computations and
expensive cosmological simulations to run in terms of time and computing
power. In the present work, we build and train a GAN to look further
into the strengths and limitations of such an approach. We then propose
a novel method in which we make use of a trained GAN to construct a
simple autoencoder (AE) as a first step towards building a predictive
model. Both the GAN and AE are trained on images issued from two types
of N-body simulations, namely 2D and 3D simulations. We find that the
GAN successfully generates new images that are statistically consistent
with the images it was trained on. We then show that the AE manages to
efficiently extract information from simulation images, satisfyingly
inferring the latent encoding of the GAN to generate an image with
similar large scale structures.},
	urldate = {2020-11-13},
	journal = {arXiv e-prints},
	author = {Ullmo, Marion and Decelle, Aurélien and Aghanim, Nabila},
	month = nov,
	year = {2020},
	keywords = {Astrophysics - Cosmology and Nongalactic Astrophysics, Computer Science - Machine Learning},
	pages = {arXiv:2011.05244},
}

Downloads: 0

{"_id":"35gKpMvNmKGxcZYWy","bibbaseid":"ullmo-decelle-aghanim-encodinglargescalecosmologicalstructurewithgenerativeadversarialnetworks-2020","author_short":["Ullmo, M.","Decelle, A.","Aghanim, N."],"bibdata":{"bibtype":"article","type":"article","title":"Encoding large scale cosmological structure with Generative Adversarial Networks","volume":"2011","url":"http://adsabs.harvard.edu/abs/2020arXiv201105244U","abstract":"Recently a type of neural networks called Generative Adversarial Networks (GANs) has been proposed as a solution for fast generation of simulation-like datasets, in an attempt to bypass heavy computations and expensive cosmological simulations to run in terms of time and computing power. In the present work, we build and train a GAN to look further into the strengths and limitations of such an approach. We then propose a novel method in which we make use of a trained GAN to construct a simple autoencoder (AE) as a first step towards building a predictive model. Both the GAN and AE are trained on images issued from two types of N-body simulations, namely 2D and 3D simulations. We find that the GAN successfully generates new images that are statistically consistent with the images it was trained on. We then show that the AE manages to efficiently extract information from simulation images, satisfyingly inferring the latent encoding of the GAN to generate an image with similar large scale structures.","urldate":"2020-11-13","journal":"arXiv e-prints","author":[{"propositions":[],"lastnames":["Ullmo"],"firstnames":["Marion"],"suffixes":[]},{"propositions":[],"lastnames":["Decelle"],"firstnames":["Aurélien"],"suffixes":[]},{"propositions":[],"lastnames":["Aghanim"],"firstnames":["Nabila"],"suffixes":[]}],"month":"November","year":"2020","keywords":"Astrophysics - Cosmology and Nongalactic Astrophysics, Computer Science - Machine Learning","pages":"arXiv:2011.05244","bibtex":"@article{ullmo_encoding_2020,\n\ttitle = {Encoding large scale cosmological structure with {Generative} {Adversarial} {Networks}},\n\tvolume = {2011},\n\turl = {http://adsabs.harvard.edu/abs/2020arXiv201105244U},\n\tabstract = {Recently a type of neural networks called Generative Adversarial \nNetworks (GANs) has been proposed as a solution for fast generation of\nsimulation-like datasets, in an attempt to bypass heavy computations and\nexpensive cosmological simulations to run in terms of time and computing\npower. In the present work, we build and train a GAN to look further\ninto the strengths and limitations of such an approach. We then propose\na novel method in which we make use of a trained GAN to construct a\nsimple autoencoder (AE) as a first step towards building a predictive\nmodel. Both the GAN and AE are trained on images issued from two types\nof N-body simulations, namely 2D and 3D simulations. We find that the\nGAN successfully generates new images that are statistically consistent\nwith the images it was trained on. We then show that the AE manages to\nefficiently extract information from simulation images, satisfyingly\ninferring the latent encoding of the GAN to generate an image with\nsimilar large scale structures.},\n\turldate = {2020-11-13},\n\tjournal = {arXiv e-prints},\n\tauthor = {Ullmo, Marion and Decelle, Aurélien and Aghanim, Nabila},\n\tmonth = nov,\n\tyear = {2020},\n\tkeywords = {Astrophysics - Cosmology and Nongalactic Astrophysics, Computer Science - Machine Learning},\n\tpages = {arXiv:2011.05244},\n}\n\n","author_short":["Ullmo, M.","Decelle, A.","Aghanim, N."],"key":"ullmo_encoding_2020","id":"ullmo_encoding_2020","bibbaseid":"ullmo-decelle-aghanim-encodinglargescalecosmologicalstructurewithgenerativeadversarialnetworks-2020","role":"author","urls":{"Paper":"http://adsabs.harvard.edu/abs/2020arXiv201105244U"},"keyword":["Astrophysics - Cosmology and Nongalactic Astrophysics","Computer Science - Machine Learning"],"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://bibbase.org/zotero/polyphant","dataSources":["7gvjSdWrEu7z5vjjj"],"keywords":["astrophysics - cosmology and nongalactic astrophysics","computer science - machine learning"],"search_terms":["encoding","large","scale","cosmological","structure","generative","adversarial","networks","ullmo","decelle","aghanim"],"title":"Encoding large scale cosmological structure with Generative Adversarial Networks","year":2020}