Constraining cosmological parameters from N-body simulations with Variational Bayesian Neural Networks
Authors:
Héctor J. Hortúa,
Luz Ángela García,
Leonardo Castañeda C
Abstract:
Methods based on Deep Learning have recently been applied on astrophysical parameter recovery thanks to their ability to capture information from complex data. One of these methods is the approximate Bayesian Neural Networks (BNNs) which have demonstrated to yield consistent posterior distribution into the parameter space, helpful for uncertainty quantification. However, as any modern neural netwo…
▽ More
Methods based on Deep Learning have recently been applied on astrophysical parameter recovery thanks to their ability to capture information from complex data. One of these methods is the approximate Bayesian Neural Networks (BNNs) which have demonstrated to yield consistent posterior distribution into the parameter space, helpful for uncertainty quantification. However, as any modern neural networks, they tend to produce overly confident uncertainty estimates and can introduce bias when BNNs are applied to data. In this work, we implement multiplicative normalizing flows (MNFs), a family of approximate posteriors for the parameters of BNNs with the purpose of enhancing the flexibility of the variational posterior distribution, to extract $Ω_m$, $h$, and $σ_8$ from the QUIJOTE simulations. We have compared this method with respect to the standard BNNs, and the flipout estimator. We found that MNFs combined with BNNs outperform the other models obtaining predictive performance with almost one order of magnitude larger that standard BNNs, $σ_8$ extracted with high accuracy ($r^2=0.99$), and precise uncertainty estimates. The latter implies that MNFs provide more realistic predictive distribution closer to the true posterior mitigating the bias introduced by the variational approximation and allowing to work with well-calibrated networks.
△ Less
Submitted 9 January, 2023;
originally announced January 2023.
Space Photometry with BRITE-Constellation
Authors:
Weiss W. W,
Zwintz K.,
Kuschnig R.,
Handler G.,
Moffat A. F. J.,
Baade D.,
Bowman D. M.,
Granzer T.,
Kallinger T.,
Koudelka O. F.,
Lovekin C. C.,
Neiner C.,
Pablo H.,
Pigulski A.,
Popowicz A.,
Ramiaramanantsoa T.,
Rucinski S. M.,
Strassmeier K. G.,
Wade G. A
Abstract:
BRITE-Constellation is devoted to high-precision optical photometric monitoring of bright stars, distributed all over the Milky Way, in red and/or blue passbands. Photometry from space avoids the turbulent and absorbing terrestrial atmosphere and allows for very long and continuous observing runs with high time resolution and thus provides the data necessary for understanding various processes ins…
▽ More
BRITE-Constellation is devoted to high-precision optical photometric monitoring of bright stars, distributed all over the Milky Way, in red and/or blue passbands. Photometry from space avoids the turbulent and absorbing terrestrial atmosphere and allows for very long and continuous observing runs with high time resolution and thus provides the data necessary for understanding various processes inside stars (e.g., asteroseismology) and in their immediate environment. While the first astronomical observations from space focused on the spectral regions not accessible from the ground it soon became obvious around 1970 that avoiding the turbulent terrestrial atmosphere significantly improved the accuracy of photometry and satellites explicitly dedicated to high-quality photometry were launched. A perfect example is BRITE-Constellation, which is the result of a very successful cooperation between Austria, Canada and Poland. Research highlights for targets distributed nearly over the entire HRD are presented, but focus primarily on massive and hot stars.
△ Less
Submitted 24 June, 2021;
originally announced June 2021.
