BayesFlow

• Verfasst von: Radev, Stefan [VerfasserIn]
• Verfasst von: Mertens, Ulf K. [VerfasserIn]
• Verfasst von: Voß, Andreas [VerfasserIn]
• Verfasst von: Ardizzone, Lynton [VerfasserIn]
• Verfasst von: Köthe, Ullrich [VerfasserIn]
• Titel: BayesFlow
• Titelzusatz: learning complex stochastic models with invertible neural networks
• Verf.angabe: Stefan T. Radev, Ulf K. Mertens, Andreas Voss, Lynton Ardizzone, and Ullrich Köthe, Member, IEEE
• Jahr: 2022
• Umfang: 15 S.
• Fussnoten: Date of publication: 18 December 2020 ; Gesehen am 31.05.2022
• Titel Quelle: Enthalten in: Institute of Electrical and Electronics EngineersIEEE transactions on neural networks and learning systems
• Ort Quelle: [New York, NY] : IEEE, 2012
• Jahr Quelle: 2022
• Band/Heft Quelle: 33(2022), 4, Seite 1452-1466
• ISSN Quelle: 2162-2388
• DOI: doi:10.1109/TNNLS.2020.3042395
• Datenträger: Online-Ressource
• Sprache: eng
• Sach-SW: Bayes methods
• Sach-SW: Bayesian inference
• Sach-SW: Biological system modeling
• Sach-SW: computational and artificial intelligence
• Sach-SW: Data models
• Sach-SW: Estimation
• Sach-SW: machine learning
• Sach-SW: neural networks
• Sach-SW: Neural networks
• Sach-SW: Numerical models
• Sach-SW: statistical learning
• Sach-SW: Training
• K10plus-PPN: 1801173818

Abstract

Estimating the parameters of mathematical models is a common problem in almost all branches of science. However, this problem can prove notably difficult when processes and model descriptions become increasingly complex and an explicit likelihood function is not available. With this work, we propose a novel method for globally amortized Bayesian inference based on invertible neural networks that we call BayesFlow. The method uses simulations to learn a global estimator for the probabilistic mapping from observed data to underlying model parameters. A neural network pretrained in this way can then, without additional training or optimization, infer full posteriors on arbitrarily many real data sets involving the same model family. In addition, our method incorporates a summary network trained to embed the observed data into maximally informative summary statistics. Learning summary statistics from data makes the method applicable to modeling scenarios where standard inference techniques with handcrafted summary statistics fail. We demonstrate the utility of BayesFlow on challenging intractable models from population dynamics, epidemiology, cognitive science, and ecology. We argue that BayesFlow provides a general framework for building amortized Bayesian parameter estimation machines for any forward model from which data can be simulated.