| 1. |
- Alsing, Justin, et al.
(författare)
-
Fast likelihood-free cosmology with neural density estimators and active learning
- 2019
-
Ingår i: Monthly notices of the Royal Astronomical Society. - : Oxford University Press (OUP). - 0035-8711 .- 1365-2966. ; 488:3, s. 4440-4458
-
Tidskriftsartikel (refereegranskat)abstract
- Likelihood-free inference provides a framework for performing rigorous Bayesian inference using only forward simulations, properly accounting for all physical and observational effects that can be successfully included in the simulations, The key challenge for likelihood-free applications in cosmology, where simulation is typically expensive, is developing methods that can achieve high-fidelity posterior inference with as few simulations as possible. Density estimation likelihood-free inference (DELFI) methods turn inference into a density-estimation task on a set of simulated data-parameter pairs, and give orders of magnitude improvements over traditional Approximate Bayesian Computation approaches to likelihood-free inference. In this paper, we use neural density estimators (NDEs) to learn the likelihood function from a set of simulated data sets, with active learning to adaptively acquire simulations in the most relevant regions of parameter space on the fly. We demonstrate the approach on a number of cosmological case studies, showing that for typical problems high-fidelity posterior inference can be achieved with just 0(103) simulations or fewer. In addition to enabling efficient simulation-based inference, for simple problems where the form of the likelihood is known, DELFI offers a fast alternative to Markov Chain Monte Carlo (MCMC) sampling, giving orders of magnitude speed-up in some cases. Finally, we introduce PYDELFI a flexible public implementation of DELFI with NDFs and active learning - available at haps://github.com/justinalsing/pydelfi.
|
|
| 2. |
- Alsing, Justin, et al.
(författare)
-
Forward Modeling of Galaxy Populations for Cosmological Redshift Distribution Inference
- 2023
-
Ingår i: Astrophysical Journal Supplement Series. - : American Astronomical Society. - 0067-0049 .- 1538-4365. ; 264:2
-
Tidskriftsartikel (refereegranskat)abstract
- We present a forward-modeling framework for estimating galaxy redshift distributions from photometric surveys. Our forward model is composed of: a detailed population model describing the intrinsic distribution of the physical characteristics of galaxies, encoding galaxy evolution physics; a stellar population synthesis model connecting the physical properties of galaxies to their photometry; a data model characterizing the observation and calibration processes for a given survey; and explicit treatment of selection cuts, both into the main analysis sample and for the subsequent sorting into tomographic redshift bins. This approach has the appeal that it does not rely on spectroscopic calibration data, provides explicit control over modeling assumptions and builds a direct bridge between photo-z inference and galaxy evolution physics. In addition to redshift distributions, forward modeling provides a framework for drawing robust inferences about the statistical properties of the galaxy population more generally. We demonstrate the utility of forward modeling by estimating the redshift distributions for the Galaxy And Mass Assembly (GAMA) survey and the Vimos VLT Deep Survey (VVDS), validating against their spectroscopic redshifts. Our baseline model is able to predict tomographic redshift distributions for GAMA and VVDS with respective biases of Δz ≲ 0.003 and Δz ≃ 0.01 on the mean redshift—comfortably accurate enough for Stage III cosmological surveys—without any hyperparameter tuning (i.e., prior to doing any fitting to those data). We anticipate that with additional hyperparameter fitting and modeling improvements, forward modeling will provide a path to accurate redshift distribution inference for Stage IV surveys.
|
|
| 3. |
- Alsing, Justin, et al.
(författare)
-
Nested sampling with any prior you like
- 2021
-
Ingår i: Monthly Notices of the Royal Astronomical Society: Letters. - : Oxford University Press (OUP). - 1745-3925 .- 1745-3933. ; 505:1, s. L95-L99
-
Tidskriftsartikel (refereegranskat)abstract
- Nested sampling is an important tool for conducting Bayesian analysis in Astronomy and other fields, both for sampling complicated posterior distributions for parameter inference, and for computing marginal likelihoods for model comparison. One technical obstacle to using nested sampling in practice is the requirement (for most common implementations) that prior distributions be provided in the form of transformations from the unit hyper-cube to the target prior density. For many applications - particularly when using the posterior from one experiment as the prior for another - such a transformation is not readily available. In this letter, we show that parametric bijectors trained on samples from a desired prior density provide a general purpose method for constructing transformations from the uniform base density to a target prior, enabling the practical use of nested sampling under arbitrary priors. We demonstrate the use of trained bijectors in conjunction with nested sampling on a number of examples from cosmology.
|
|
| 4. |
- Alsing, Justin, et al.
(författare)
-
Nuisance hardened data compression for fast likelihood-free inference
- 2019
-
Ingår i: Monthly notices of the Royal Astronomical Society. - : Oxford University Press (OUP). - 0035-8711 .- 1365-2966. ; 488:4, s. 5093-5103
-
Tidskriftsartikel (refereegranskat)abstract
- We show how nuisance parameter marginalized posteriors can be inferred directly from simulations in a likelihood-free setting, without having to jointly infer the higher dimensional interesting and nuisance parameter posterior first and marginalize a posteriori. The result is that for an inference task with a given number of interesting parameters, the number of simulations required to perform likelihood-free inference can be kept (roughly) the same irrespective of the number of additional nuisances to be marginalized over. To achieve this, we introduce two extensions to the standard likelihood-free inference set-up. First, we show how nuisance parameters can be recast as latent variables and hence automatically marginalized over in the likelihood-free framework. Secondly, we derive an asymptotically optimal compression from N data to n summaries - one per interesting parameter - such that the Fisher information is (asymptotically) preserved, but the summaries are insensitive to the nuisance parameters. This means that the nuisance marginalized inference task involves learning n interesting parameters from n nuisance hardened' data summaries, regardless of the presence or number of additional nuisance parameters to be marginalized over. We validate our approach on two examples from cosmology: supernovae and weak-lensing data analyses with nuisance parametrized systematics. For the supernova problem, high-fidelity posterior inference of Omega(m) and w(0) (marginalized over systematics) can be obtained from just a few hundred data simulations. For the weak-lensing problem, six cosmological parameters can be inferred from just simulations, irrespective of whether 10 additional nuisance parameters are included in the problem or not.
|
|
| 5. |
- Alsing, Justin, et al.
(författare)
-
SPECULATOR : Emulating Stellar Population Synthesis for Fast and Accurate Galaxy Spectra and Photometry
- 2020
-
Ingår i: Astrophysical Journal Supplement Series. - : American Astronomical Society. - 0067-0049 .- 1538-4365. ; 249:1
-
Tidskriftsartikel (refereegranskat)abstract
- We presentspeculator-a fast, accurate, and flexible framework for emulating stellar population synthesis (SPS) models for predicting galaxy spectra and photometry. For emulating spectra, we use a principal component analysis to construct a set of basis functions and neural networks to learn the basis coefficients as a function of the SPS model parameters. For photometry, we parameterize the magnitudes (for the filters of interest) as a function of SPS parameters by a neural network. The resulting emulators are able to predict spectra and photometry under both simple and complicated SPS model parameterizations to percent-level accuracy, giving a factor of 10(3)-10(4)speedup over direct SPS computation. They have readily computable derivatives, making them amenable to gradient-based inference and optimization methods. The emulators are also straightforward to call from a GPU, giving an additional order of magnitude speedup. Rapid SPS computations delivered by emulation offers a massive reduction in the computational resources required to infer the physical properties of galaxies from observed spectra or photometry and simulate galaxy populations under SPS models, while maintaining the accuracy required for a range of applications.
|
|
| 6. |
- Ball, William T., et al.
(författare)
-
Evidence for a continuous decline in lower stratospheric ozone offsetting ozone layer recovery
- 2018
-
Ingår i: Atmospheric Chemistry And Physics. - : Copernicus GmbH. - 1680-7316 .- 1680-7324. ; 18:2, s. 1379-1394
-
Tidskriftsartikel (refereegranskat)abstract
- Ozone forms in the Earth's atmosphere from the photodissociation of molecular oxygen, primarily in the tropical stratosphere. It is then transported to the extratropics by the Brewer-Dobson circulation (BDC), forming a protective ozone layer around the globe. Human emissions of halogen-containing ozone-depleting substances (hODSs) led to a decline in stratospheric ozone until they were banned by the Montreal Protocol, and since 1998 ozone in the upper stratosphere is rising again, likely the recovery from halogen-induced losses. Total column measurements of ozone between the Earth's surface and the top of the atmosphere indicate that the ozone layer has stopped declining across the globe, but no clear increase has been observed at latitudes between 60 degrees S and 60 degrees N outside the polar regions (60-90 degrees). Here we report evidence from multiple satellite measurements that ozone in the lower stratosphere between 60 degrees S and 60 degrees N has indeed continued to decline since 1998. We find that, even though upper stratospheric ozone is recovering, the continuing downward trend in the lower stratosphere prevails, resulting in a downward trend in stratospheric column ozone between 60 degrees S and 60 degrees N. We find that total column ozone between 60 degrees S and 60 degrees N appears not to have decreased only because of increases in tropospheric column ozone that compensate for the stratospheric decreases. The reasons for the continued reduction of lower stratospheric ozone are not clear; models do not reproduce these trends, and thus the causes now urgently need to be established.
|
|
| 7. |
- Ball, William T., et al.
(författare)
-
Inconsistencies between chemistry-climate models and observed lower stratospheric ozone trends since 1998
- 2020
-
Ingår i: Atmospheric Chemistry And Physics. - : Copernicus GmbH. - 1680-7316 .- 1680-7324. ; 20:16, s. 9737-9752
-
Tidskriftsartikel (refereegranskat)abstract
- The stratospheric ozone layer shields surface life from harmful ultraviolet radiation. Following the Montreal Protocol ban on long-lived ozone-depleting substances (ODSs), rapid depletion of total column ozone (TCO) ceased in the late 1990s, and ozone above 32 km is now clearly recovering. However, there is still no confirmation of TCO recovery, and evidence has emerged that ongoing quasiglobal (60 degrees S-60 degrees N) lower stratospheric ozone decreases may be responsible, dominated by low latitudes (30 degrees S-30 degrees N). Chemistry-climate models (CCMs) used to project future changes predict that lower stratospheric ozone will decrease in the tropics by 2100 but not at mid-latitudes (30-60 degrees). Here, we show that CCMs display an ozone decline similar to that observed in the tropics over 1998-2016, likely driven by an increase in tropical upwelling. On the other hand, mid-latitude lower stratospheric ozone is observed to decrease, while CCMs that specify real-world historical meteorological fields instead show an increase up to present day. However, these cannot be used to simulate future changes; we demonstrate here that free-running CCMs used for projections also show increases. Despite opposing lower stratospheric ozone changes, which should induce opposite temperature trends, CCMs and observed temperature trends agree; we demonstrate that opposing model- observation stratospheric water vapour (SWV) trends, and their associated radiative effects, explain why temperature changes agree in spite of opposing ozone trends. We provide new evidence that the observed mid-latitude trends can be explained by enhanced mixing between the tropics and extratropics. We further show that the temperature trends are consistent with the observed mid-latitude ozone decrease. Together, our results suggest that large-scale circulation changes expected in the future from increased greenhouse gases (GHGs) may now already be underway but that most CCMs do not simulate mid-latitude ozone layer changes well. However, it is important to emphasise that the periods considered here are short, and internal variability that is both intrinsic to each CCM and different to observed historical variability is not well-characterised and can influence trend estimates. Nevertheless, the reason CCMs do not exhibit the observed changes needs to be identified to allow models to be improved in order to build confidence in future projections of the ozone layer.
|
|
| 8. |
- Ball, William T., et al.
(författare)
-
Reconciling differences in stratospheric ozone composites
- 2017
-
Ingår i: Atmospheric Chemistry And Physics. - : Copernicus GmbH. - 1680-7316 .- 1680-7324. ; 17:20, s. 12269-12302
-
Tidskriftsartikel (refereegranskat)abstract
- Observations of stratospheric ozone from multiple instruments now span three decades; combining these into composite datasets allows long-term ozone trends to be estimated. Recently, several ozone composites have been published, but trends disagree by latitude and altitude, even between composites built upon the same instrument data. We confirm that the main causes of differences in decadal trend estimates lie in (i) steps in the composite time series when the instrument source data changes and (ii) artificial sub-decadal trends in the underlying instrument data. These artefacts introduce features that can alias with regressors in multiple linear regression (MLR) analysis; both can lead to inaccurate trend estimates. Here, we aim to remove these artefacts using Bayesian methods to infer the underlying ozone time series from a set of composites by building a joint-likelihood function using a Gaussian-mixture density to model outliers introduced by data artefacts, together with a data-driven prior on ozone variability that incorporates knowledge of problems during instrument operation. We apply this Bayesian self-calibration approach to stratospheric ozone in 10 degrees bands from 60 degrees S to 60 degrees N and from 46 to 1 hPa (similar to 21-48 km) for 1985-2012. There are two main outcomes: (i) we independently identify and confirm many of the data problems previously identified, but which remain unaccounted for in existing composites; (ii) we construct an ozone composite, with uncertainties, that is free from most of these problems - we call this the BAyeSian Integrated and Consolidated (BASIC) composite. To analyse the new BASIC composite, we use dynamical linear modelling (DLM), which provides a more robust estimate of long-term changes through Bayesian inference than MLR. BASIC and DLM, together, provide a step forward in improving estimates of decadal trends. Our results indicate a significant recovery of ozone since 1998 in the upper stratosphere, of both northern and southern midlatitudes, in all four composites analysed, and particularly in the BASIC composite. The BASIC results also show no hemispheric difference in the recovery at midlatitudes, in contrast to an apparent feature that is present, but not consistent, in the four composites. Our overall conclusion is that it is possible to effectively combine different ozone composites and account for artefacts and drifts, and that this leads to a clear and significant result that upper stratospheric ozone levels have increased since 1998, following an earlier decline.
|
|
| 9. |
- Ball, William T., et al.
(författare)
-
Stratospheric ozone trends for 1985-2018 : sensitivity to recent large variability
- 2019
-
Ingår i: Atmospheric Chemistry And Physics. - : Copernicus GmbH. - 1680-7316 .- 1680-7324. ; 19:19, s. 12731-12748
-
Tidskriftsartikel (refereegranskat)abstract
- The Montreal Protocol, and its subsequent amendments, has successfully prevented catastrophic losses of stratospheric ozone, and signs of recovery are now evident. Nevertheless, recent work has suggested that ozone in the lower stratosphere (< 24 km) continued to decline over the 1998-2016 period, offsetting recovery at higher altitudes and preventing a statistically significant increase in quasi-global (60 degrees S-60 degrees N) total column ozone. In 2017, a large lower stratospheric ozone resurgence over less than 12 months was estimated (using a chemistry transport model; CTM) to have offset the long-term decline in the quasi-global integrated lower stratospheric ozone column. Here, we extend the analysis of space-based ozone observations to December 2018 using the BASIC(SG) ozone composite. We find that the observed 2017 resurgence was only around half that modelled by the CTM, was of comparable magnitude to other strong interannual changes in the past, and was restricted to Southern Hemisphere (SH) midlatitudes (60-30 degrees S). In the SH midlatitude lower stratosphere, the data suggest that by the end of 2018 ozone is still likely lower than in 1998 (probability similar to 80 %). In contrast, tropical and Northern Hemisphere (NH) ozone continue to display ongoing decreases, exceeding 90 % probability. Robust tropical (> 95 %, 30 degrees S-30 degrees N) decreases dominate the quasi-global integrated decrease (99 % probability); the integrated tropical stratospheric column (1-100 hPa, 30 degrees S-30 degrees N) displays a significant overall ozone decrease, with 95 % probability. These decreases do not reveal an inefficacy of the Montreal Protocol; rather, they suggest that other effects are at work, mainly dynamical variability on long or short timescales, counteracting the positive effects of the Montreal Protocol on stratospheric ozone recovery. We demonstrate that large interannual midlatitude (30-60 degrees) variations, such as the 2017 resurgence, are driven by non-linear quasi-biennial oscillation (QBO) phase-dependent seasonal variability. However, this variability is not represented in current regression analyses. To understand if observed lower stratospheric ozone decreases are a transient or long-term phenomenon, progress needs to be made in accounting for this dynamically driven variability.
|
|
| 10. |
- Bevins, Harry T. J., et al.
(författare)
-
Marginal post-processing of Bayesian inference products with normalizing flows and kernel density estimators
- 2023
-
Ingår i: Monthly notices of the Royal Astronomical Society. - 0035-8711 .- 1365-2966. ; 526:3, s. 4613-4626
-
Tidskriftsartikel (refereegranskat)abstract
- Bayesian analysis has become an indispensable tool across many different cosmological fields, including the study of gravitational waves, the cosmic microwave background, and the 21-cm signal from the Cosmic Dawn, among other phenomena. The method provides a way to fit complex models to data describing key cosmological and astrophysical signals and a whole host of contaminating signals and instrumental effects modelled with ‘nuisance parameters’. In this paper, we summarize a method that uses masked autoregressive flows and kernel density estimators to learn marginal posterior densities corresponding to core science parameters. We find that the marginal or ‘nuisance-free’ posteriors and the associated likelihoods have an abundance of applications, including the calculation of previously intractable marginal Kullback–Leibler divergences and marginal Bayesian model dimensionalities, likelihood emulation, and prior emulation. We demonstrate each application using toy examples, examples from the field of 21-cm cosmology, and samples from the Dark Energy Survey. We discuss how marginal summary statistics like the Kullback–Leibler divergences and Bayesian model dimensionalities can be used to examine the constraining power of different experiments and how we can perform efficient joint analysis by taking advantage of marginal prior and likelihood emulators. We package our multipurpose code up in the pip-installable code MARGARINE for use in the wider scientific community.
|
|
