swyft: Truncated Marginal Neural Ratio Estimation in
Python

Miller, Benjamin Kurt; Cole, Alex; Weniger, Christoph; Nattino, Francesco; Ku, Ou; Grootes, Meiert W.

doi:10.21105/joss.04205

Cited by 21 publications

(21 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The tradeoff is that the likelihood for our persistence diagrams is only implicitly defined. Parameter estimation in the context of implicit likelihoods is precisely within the purview of the rapidly advancing field of simulation-based inference ( [83][84][85][86], see [87] for a recent review and [88][89][90][91][92] for cosmological applications). Within simulation-based inference it has been advocated (see e.g.…”

Section: Discussionmentioning

confidence: 99%

Fisher forecasts for primordial non-Gaussianity from persistent homology

Biagetti

Calles²,

Castiblanco³

et al. 2022

J. Cosmol. Astropart. Phys.

Self Cite

View full text Add to dashboard Cite

We study the information content of summary statistics built from the multi-scale topology of large-scale structures on primordial non-Gaussianity of the local and equilateral type. We use halo catalogs generated from numerical N-body simulations of the Universe on large scales as a proxy for observed galaxies. Besides calculating the Fisher matrix for halos in real space, we also check more realistic scenarios in redshift space. Without needing to take a distant observer approximation, we place the observer on a corner of the box. We also add redshift errors mimicking spectroscopic and photometric samples. We perform several tests to assess the reliability of our Fisher matrix, including the Gaussianity of our summary statistics and convergence. We find that the marginalized 1-σ uncertainties in redshift space are Δf NL loc ∼ 16 and Δf NL equi ∼ 41 on a survey volume of 1 (Gpc/h)3. These constraints are weakly affected by redshift errors. We close by speculating as to how this approach can be made robust against small-scale uncertainties by exploiting (non)locality.

show abstract

Section: Discussionmentioning

confidence: 99%

Fisher forecasts for primordial non-Gaussianity from persistent homology

Biagetti

Calles²,

Castiblanco³

et al. 2022

J. Cosmol. Astropart. Phys.

Self Cite

View full text Add to dashboard Cite

show abstract

“…In particular, it might be interesting to employ a graph convolutional neural network as suggested in [41], in order to preserve spatial relations between distant pixels, and rotational invariance. Finally, recent developments in the field of simulation based inference, such as our case, suggest that a promising approach to estimate the complex posterior distribution of the objective parameters are the so-called neural likelihood ratio estimation techniques [60], and in particular Truncated Marginal Neural Ratio Estimation [61][62][63]. Since these techniques are tailored for simulation based problems, we hypothesise that it would be possible to achieve similar performance to our current neural network architecture with an even simpler structure and possibly less training samples.…”

Section: Jcap09(2023)029mentioning

confidence: 99%

Extracting the gamma-ray source-count distribution below the Fermi-LAT detection limit with deep learning

Amerio,

Cuoco,

Fornengo

2023

J. Cosmol. Astropart. Phys.

View full text Add to dashboard Cite

We reconstruct the extra-galactic gamma-ray source-count distribution, or dN/dS, of resolved and unresolved sources by adopting machine learning techniques. Specifically, we train a convolutional neural network on synthetic 2-dimensional sky-maps, which are built by varying parameters of underlying source-counts models and incorporate the Fermi-LAT instrumental response functions. The trained neural network is then applied to the Fermi-LAT data, from which we estimate the source count distribution down to flux levels a factor of 50 below the Fermi-LAT threshold. We perform our analysis using 14 years of data collected in the (1,10) GeV energy range. The results we obtain show a source count distribution which, in the resolved regime, is in excellent agreement with the one derived from cataloged sources, and then extends as dN/dS ∼ S -2 in the unresolved regime, down to fluxes of 5 · 10-12 cm-2 s-1. The neural network architecture and the devised methodology have the flexibility to enable future analyses to study the energy dependence of the source-count distribution.

show abstract

“…When allocated one node (32 CPU threads) on highperformance computing clusters, our current nested sampling infrastructure sees typical per-target timescales on the order of a week. As such, our future work will instead rely on the development of a simulation-based inference (SBI; Cranmer et al 2020) machine-learning infrastructure; these have seen great success in recent years (see Alsing et al 2018Alsing et al , 2019Miller et al 2020;Tejero-Cantero et al 2020;Miller et al 2022;Legin et al 2023b). The amortized nature of SBI will allow for computationally efficient deployment across parameter space in catalog-wide applications to current and future missions (Kepler,K2,TESS,PLATO,etc.…”

Section: Next Stepsmentioning

confidence: 99%

Gaussian Processes and Nested Sampling Applied to Kepler's Small Long-period Exoplanet Candidates

Matesic,

Rowe,

Livingston

et al. 2024

View full text Add to dashboard Cite

There are more than 5000 confirmed and validated planets beyond the solar system to date, more than half of which were discovered by NASA’s Kepler mission. The catalog of Kepler’s exoplanet candidates has only been extensively analyzed under the assumption of white noise (i.i.d. Gaussian), which breaks down on timescales longer than a day due to correlated noise (point-to-point correlation) from stellar variability and instrumental effects. Statistical validation of candidate transit events becomes increasingly difficult when they are contaminated by this form of correlated noise, especially in the low-signal-to-noise (S/N) regimes occupied by Earth–Sun and Venus–Sun analogs. To diagnose small long-period, low-S/N putative transit signatures with few (roughly 3–9) observed transit-like events (e.g., Earth–Sun analogs), we model Kepler's photometric data as noise, treated as a Gaussian process, with and without the inclusion of a transit model. Nested sampling algorithms from the Python UltraNest package recover model evidences and maximum a posteriori parameter sets, allowing us to disposition transit signatures as either planet candidates or false alarms within a Bayesian framework.

show abstract

swyft: Truncated Marginal Neural Ratio Estimation in Python

Cited by 21 publications

References 35 publications

Fisher forecasts for primordial non-Gaussianity from persistent homology

Fisher forecasts for primordial non-Gaussianity from persistent homology

Extracting the gamma-ray source-count distribution below the Fermi-LAT detection limit with deep learning

Gaussian Processes and Nested Sampling Applied to Kepler's Small Long-period Exoplanet Candidates

Contact Info

Product

Resources

About