\pkg BayesFlow 2.0: Multi-Backend Amortized Bayesian Inference in \proglang Python

Lars Kühmichel TU Dortmund University, Jerry M. Huang Rensselaer Polytechnic Institute, Valentin Pratz Heidelberg University, Jonas Arruda University of Bonn, Hans Olischläger TU Dortmund University, Daniel Habermann TU Dortmund University, Šimon Kucharský TU Dortmund University, Lasse Elsemüller In · 2026

Read \pkg BayesFlow 2.0: Multi-Backend Amortized Bayesian Inference in \proglang Python by Lars Kühmichel TU Dortmund University, Jerry M. Huang Rensselaer Polytechnic Institute, Valentin Pratz Heidelberg University, Jonas Arruda University of Bonn, Hans Olischläger TU Dortmund University, Daniel Habermann TU Dortmund University, Šimon Kucharský TU Dortmund University, Lasse Elsemüller In (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07098.txt.

Description

Lars Kühmichel and thirteen co-authors offer BayesFlow 2.0, a Python library for amortized Bayesian inference. Modern Bayesian work mixes many computational methods. Those methods figure out, validate, and draw conclusions from probabilistic models, yet they run slowly. Slow fitting often blocks complex models on large data sets. Amortized inference trains neural networks on model simulations, then rewards users with fast answers for any quantity the model implies, from point figures and likelihoods to full posterior distributions.

Version 2.0 covers direct posterior, likelihood, and ratio figuring. It supports multiple popular deep learning backends, with a rich set of networks that make samples and gauge densities, full customization, and high-level interfaces. New features add tuning of settings, design tuning, and layered modeling. A test case that fits parameters of a dynamical system, plus comparisons with similar software, shows a streamlined, user-friendly workflow with strong potential for broad adoption.

Catalog record

CREATORLars Kühmichel TU Dortmund University, Jerry M. Huang Rensselaer Polytechnic Institute, Valentin Pratz Heidelberg University, Jonas Arruda University of Bonn, Hans Olischläger TU Dortmund University, Daniel Habermann TU Dortmund University, Šimon Kucharský TU Dortmund University, Lasse Elsemüller In
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07098

Browse the full library · About the Tower of Babel