Real-time Likelihood-free Inference of Roman Binary Microlensing Events with Amortized Neural Posterior Estimation

Abstract

Fast and automated inference of binary-lens, single-source (2L1S) microlensing events with sampling-based Bayesian algorithms (e.g., Markov Chain Monte Carlo, MCMC) is challenged on two fronts: the high computational cost of likelihood evaluations with microlensing simulation codes, and a pathological parameter space where the negative-log-likelihood surface can contain a multitude of local minima that are narrow and deep. Analysis of 2L1S events usually involves grid searches over some parameters to locate approximate solutions as a prerequisite to posterior sampling, an expensive process that often requires human-in-the-loop domain expertise. As the next-generation, space-based microlensing survey with the Roman Space Telescope is expected to yield thousands of binary microlensing events, a new fast and automated method is desirable. Here, we present a likelihood-free inference approach named amortized neural posterior estimation, where a neural density estimator (NDE) learns a surrogate posterior as an observation-parameterized conditional probability distribution, from pre-computed simulations over the full prior space. Trained on 291,012 simulated Roman-like 2L1S simulations, the NDE produces accurate and precise posteriors within seconds for any observation within the prior support without requiring a domain expert in the loop, thus allowing for real-time and automated inference. We show that the NDE also captures expected posterior degeneracies. The NDE posterior could then be refined into the exact posterior with a downstream MCMC sampler with minimal burn-in steps.

Accelerated Bayesian SED Modeling Using Amortized Neural Posterior Estimation

“…Simulation-based inference (SBI) is a rapidly developing class of inference methods that offers alternatives for many applications (see Cranmer et al 2020, and references therein). Many SBI methods leverage the latest developments in statistics and machine learning for more efficient posterior estimation (Papamakarios et al 2017;Alsing et al 2019;Hahn et al 2019;Jeffrey & Wandelt 2020;Dax et al 2021;Huppenkothen & Bachetti 2021;Zhang et al 2021). Of particular interest for SED modeling is a technique called Amortized Neural Posterior Estimation (ANPE).…”

Astrophysical and Cosmological Probes of Dark Matter

“…4. Neural SBI has been used, for instance, for dark matter substructure inference [628,[645][646][647], dark matter indirect detection with gamma-ray data [648][649][650], and binary microlensing [651]. Fronts where still significant theoretical development is required are neural network architectures tailored to the structure of typical astrophysical data, which can significantly reduce simulation costs for simulation-based algorithms, and simulation-efficient training algorithms.…”

Accelerated Bayesian SED Modeling using Amortized Neural Posterior Estimation

“…SBI has already been successfully applied to a number of Bayesian parameter inference problems in astronomy (e.g. Cameron & Pettitt 2012;Weyant et al 2013;Hahn et al 2017;Kacprzak et al 2018;Alsing et al 2018;Wong et al 2020;Huppenkothen & Bachetti 2021;Zhang et al 2021) and in physics (e.g. Brehmer et al 2019;Cranmer et al 2020).…”

“…This technique is called Amortized Neural Posterior Estimation (ANPE) and has recently been applied to a broad range of astronomical applications from analyzing gravitational waves (e.g. Wong et al 2020;Dax et al 2021) to binary microlensing lensing (Zhang et al 2021). For SED modeling, the choice in favor of using ANPE is easy: the entire upfront cost for ANPE model evaluations would only yield posteriors of tens of galaxies with MCMC.…”

“…Simulation-based inference (SBI) is a rapidly developing class of inference methods that offers alternatives for many applications (see Cranmer et al 2020, and references therein). Many SBI methods leverage the latest developments in statistics and Machine Learning for more efficient posterior estimation (Papamakarios et al 2017;Alsing et al 2019;Hahn et al 2019;Dax et al 2021;Huppenkothen & Bachetti 2021;Zhang et al 2021). Of particular interest for SED modeling is a technique called Amortized Neural Posterior Estimation (ANPE).…”