A Framework for Obtaining Accurate Posteriors of Strong Gravitational Lensing Parameters with Flexible Priors and Implicit Likelihoods Using Density Estimation

Abstract

We report the application of implicit likelihood inference to the prediction of the macroparameters of strong lensing systems with neural networks. This allows us to perform deep-learning analysis of lensing systems within a well-defined Bayesian statistical framework to explicitly impose desired priors on lensing variables, obtain accurate posteriors, and guarantee convergence to the optimal posterior in the limit of perfect performance. We train neural networks to perform a regression task to produce point estimates of lensing parameters. We then interpret these estimates as compressed statistics in our inference setup and model their likelihood function using mixture density networks. We compare our results with those of approximate Bayesian neural networks, discuss their significance, and point to future directions. Based on a test set of 100,000 strong lensing simulations, our amortized model produces accurate posteriors for any arbitrary confidence interval, with a maximum percentage deviation of 1.4% at the 21.8% confidence level, without the need for any added calibration procedure. In total, inferring 100,000 different posteriors takes a day on a single GPU, showing that the method scales well to the thousands of lenses expected to be discovered by upcoming sky surveys.

Beyond Gaussian Noise: A Generalized Approach to Likelihood Analysis with Non-Gaussian Noise

“…It is important to note that because SLIC learns the distribution of noise in the entirety of the data space, it maintains the complete information content of the data during inference. Therefore, when compared to simulation-based inference methods that resort to data compression techniques in which may lose information (Cranmer et al 2020;Legin et al 2023), SLIC is expected to provide equal (if the compression is optimal) or greater precision during inference.…”

Stochastic Modeling of Star Formation Histories. III. Constraints from Physically Motivated Gaussian Processes

“…We then estimate the posterior distribution of model parameters conditioned on an ensemble of observations, with results shown in Figure 19. Using a procedure similar to Hahn & Melchior (2022) and related papers (Khullar et al 2022;Legin et al 2023;Lemos et al 2023;Wang et al 2023), we implement SBI combined with amortized neural posterior estimation to determine that ensemble observations of about 30 galaxies allow us to constrain the timescales and scatter enough to differentiate between the toy models in this paper. Adding additional systematics like variable metallicity or observational noise broadens the posteriors, but nevertheless allows us to distinguish between the toy models and obtain constraints on the timescales, as shown in Figure 20.…”

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

“…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.…”