Bayesian parameter estimation using conditional variational autoencoders for gravitational-wave astronomy

Gabbard, H. A.; Messenger, C.; Heng, I. S.; Tonolini, Francesco; Murray-Smith, Roderick

doi:10.1038/s41567-021-01425-7

Cited by 107 publications

(56 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Past studies using machine-learning techniques for amortized GW parameter inference (Gabbard et al, 2019;Chua & Vallisneri, 2020;Green & Gair, 2021;Delaunoy et al, 2020) all consider simplified problems (e.g., only a subset of parameters, a simplified posterior, or a limited treatment of detector noise). In contrast, the GNPE-based study in Dax et al (2021) is the only one to treat the full amortized parameter inference problem with accuracy matching standard methods.…”

Section: Related Workmentioning

confidence: 99%

“…Rapid amortized methods such as NPE have the potential to transform GW data analysis. However, due to the complexity and high dimensionality 5 of GW data, it has been a challenge (Gabbard et al, 2019;Chua & Vallisneri, 2020;Green & Gair, 2021;Delaunoy et al, 2020) to obtain results of comparable accuracy and completeness to classical samplers. We now show how GNPE can be used to exploit equivariances to greatly simplify the inference problem and achieve for the first time performance indistinguishable from "ground truth" stochastic samplers-at drastically reduced inference times.…”

Section: Gravitational-wave Parameter Inferencementioning

confidence: 99%

See 1 more Smart Citation

Group equivariant neural posterior estimation

Dax¹,

Green²,

Gair³

et al. 2021

Preprint

View full text Add to dashboard Cite

Simulation-based inference with conditional neural density estimators is a powerful approach to solving inverse problems in science. However, these methods typically treat the underlying forward model as a black box, with no way to exploit geometric properties such as equivariances. Equivariances are common in scientific models, however integrating them directly into expressive inference networks (such as normalizing flows) is not straightforward. We here describe an alternative method to incorporate equivariances under joint transformations of parameters and data. Our method-called group equivariant neural posterior estimation (GNPE)-is based on self-consistently standardizing the "pose" of the data while estimating the posterior over parameters. It is architecture-independent, and applies both to exact and approximate equivariances. As a real-world application, we use GNPE for amortized inference of astrophysical binary black hole systems from gravitationalwave observations. We show that GNPE achieves state-of-the-art accuracy while reducing inference times by three orders of magnitude.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Gravitational-wave Parameter Inferencementioning

confidence: 99%

Group equivariant neural posterior estimation

Dax¹,

Green²,

Gair³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…As described in recent reviews (Huerta and Zhao, 2020 ; Cuoco et al, 2021 ), AI and high performance computing (HPC) as well as edge computing have been showcased to enable gravitational wave detection with the same sensitivity than template-matching algorithms, but orders of magnitude faster and at a fraction of the computational cost. At a glance, recent AI applications for gravitational wave astrophysics includes classification or signal detection (Gabbard et al, 2018 ; George and Huerta, 2018a , b ; Dreissigacker et al, 2019 ; Fan et al, 2019 ; Miller et al, 2019 ; Rebei et al, 2019 ; Beheshtipour and Papa, 2020 ; Deighan et al, 2020 ; Dreissigacker and Prix, 2020 ; Krastev, 2020 ; Li et al, 2020a ; Schäfer et al, 2020 , 2021 ; Skliris et al, 2020 ; Wang et al, 2020 ; Gunny et al, 2021 ; Lin and Wu, 2021 ; Schäfer and Nitz, 2021 ), signal denoising and data cleaning (Shen et al, 2019 ; Ormiston et al, 2020 ; Wei and Huerta, 2020 ; Yu and Adhikari, 2021 ), regression or parameter estimation (Gabbard et al, 2019 ; Chua and Vallisneri, 2020 ; Green and Gair, 2020 ; Green et al, 2020 ; Dax et al, 2021a , b ; Shen et al, 2022 ) Khan and Huerta 1 , accelerated waveform production (Chua et al, 2019 ; Khan and Green, 2021 ), signal forecasting (Lee et al, 2021 ; Khan et al, 2022 ), and early warning systems for gravitational wave sources that include matter, such as binary neutron stars or black hole-neutron star systems (Wei and Huerta, 2021 ; Wei et al, 2021a ; Yu et al, 2021 ).…”

Section: Introductionmentioning

confidence: 99%

Inference-Optimized AI and High Performance Computing for Gravitational Wave Detection at Scale

Chaturvedi

Khan

Tian

et al. 2022

Front. Artif. Intell.

View full text Add to dashboard Cite

We introduce an ensemble of artificial intelligence models for gravitational wave detection that we trained in the Summit supercomputer using 32 nodes, equivalent to 192 NVIDIA V100 GPUs, within 2 h. Once fully trained, we optimized these models for accelerated inference using NVIDIA TensorRT. We deployed our inference-optimized AI ensemble in the ThetaGPU supercomputer at Argonne Leadership Computer Facility to conduct distributed inference. Using the entire ThetaGPU supercomputer, consisting of 20 nodes each of which has 8 NVIDIA A100 Tensor Core GPUs and 2 AMD Rome CPUs, our NVIDIA TensorRT-optimized AI ensemble processed an entire month of advanced LIGO data (including Hanford and Livingston data streams) within 50 s. Our inference-optimized AI ensemble retains the same sensitivity of traditional AI models, namely, it identifies all known binary black hole mergers previously identified in this advanced LIGO dataset and reports no misclassifications, while also providing a 3X inference speedup compared to traditional artificial intelligence models. We used time slides to quantify the performance of our AI ensemble to process up to 5 years worth of advanced LIGO data. In this synthetically enhanced dataset, our AI ensemble reports an average of one misclassification for every month of searched advanced LIGO data. We also present the receiver operating characteristic curve of our AI ensemble using this 5 year long advanced LIGO dataset. This approach provides the required tools to conduct accelerated, AI-driven gravitational wave detection at scale.

show abstract

“…This would be especially beneficial for searches that require low latency, such as the early warning of binary neutron star mergers (Baltus et al, 2021 ; Yu et al, 2021 ). Other successful usage of ML techniques in GW astronomy include the identification of various GW events (Bayley et al, 2020 ; Chan et al, 2020 ; Dreissigacker and Prix, 2020 ; Huerta et al, 2020 ; Krastev, 2020 ; Schäfer et al, 2020 ; Wong et al, 2020 ; Beheshtipour and Papa, 2021 ; Chang et al, 2021 ; Chatterjee et al, 2021 ; López et al, 2021 ; Marianer et al, 2021 ; Mishra et al, 2021 ; Saiz-Pérez et al, 2021 ; Wei and Huerta, 2021 ; Yan et al, 2021 ), source parameter estimations (Gabbard et al, 2019 ; Chatterjee et al, 2020 ; Chua and Vallisneri, 2020 ; Green et al, 2020 ; Talbot and Thrane, 2020 ; Álvares et al, 2021 ; D'Emilio et al, 2021 ; Krastev et al, 2021 ; Williams et al, 2021 ; Xia et al, 2021 ), and detector characterization (Biswas et al, 2020 ; Colgan et al, 2020 ; Cuoco et al, 2020 ; Essick et al, 2020 ; Torres-Forné et al, 2020 ; Mogushi, 2021 ; Sankarapandian and Kulis, 2021 ; Soni et al, 2021 ; Zhan et al, 2021 ). Besides GW astronomy, the usage of CNNs has led to breakthroughs in a variety of topics related to time-series forecasting and classification (e.g., Refs.…”

Section: Introductionmentioning

confidence: 99%

Nonlinear Noise Cleaning in Gravitational-Wave Detectors With Convolutional Neural Networks

Adhikari

2022

Front. Artif. Intell.

View full text Add to dashboard Cite

Currently, the sub-60 Hz sensitivity of gravitational-wave (GW) detectors like Advanced LIGO (aLIGO) is limited by the control noises from auxiliary degrees of freedom which nonlinearly couple to the main GW readout. One promising way to tackle this challenge is to perform nonlinear noise mitigation using convolutional neural networks (CNNs), which we examine in detail in this study. In many cases, the noise coupling is bilinear and can be viewed as a few fast channels' outputs modulated by some slow channels. We show that we can utilize this knowledge of the physical system and adopt an explicit “slow×fast” structure in the design of the CNN to enhance its performance of noise subtraction. We then examine the requirements in the signal-to-noise ratio (SNR) in both the target channel (i.e., the main GW readout) and in the auxiliary sensors in order to reduce the noise by at least a factor of a few. In the case of limited SNR in the target channel, we further demonstrate that the CNN can still reach a good performance if we use curriculum learning techniques, which in reality can be achieved by combining data from quiet times and those from periods with active noise injections.

show abstract

Bayesian parameter estimation using conditional variational autoencoders for gravitational-wave astronomy

Cited by 107 publications

References 43 publications

Group equivariant neural posterior estimation

Group equivariant neural posterior estimation

Inference-Optimized AI and High Performance Computing for Gravitational Wave Detection at Scale

Nonlinear Noise Cleaning in Gravitational-Wave Detectors With Convolutional Neural Networks

Contact Info

Product

Resources

About