Simulation-Efficient Double-Flow Neural Posterior Estimation for Agent-Based Models

DOI10.48550/ARXIV.2610.35442MaRDI QIDQ7408845arXivFDO

Authors Junyao Zhang, Jinglai Li, Junqi Tang

Publication date 26 October 2026

Published in arXiv

Tldr This paper proposes Double-Flow Neural Posterior Estimation (DF-NPE), a method that reduces computational cost by combining learned time-series summaries with two conditional normalizing flows for synthetic-data generation and posterior estimation.

Full work available at URL https://doi.org/10.48550/ARXIV.2610.35442

arXiv classification

cs.MA

Cites work

• A comparative review of dimension reduction methods in approximate Bayesian computation
• A kernel two-sample test
• Adam: A Method for Stochastic Optimization
• Adaptive approximate Bayesian computation
• Agent-based model calibration using machine learning surrogates
• Approximate Bayesian Computation with the Wasserstein Distance
• Bayesian estimation of agent-based models
• Bayesian estimation of large-scale simulation models with Gaussian process regression surrogates
• Estimation of agent-based models using Bayesian deep learning approach of BayesFlow
• Estimation of agent-based models using sequential Monte Carlo methods
• Heterogeneous beliefs and routes to chaos in a simple asset pricing model
• Learning Summary Statistic for Approximate Bayesian Computation via Deep Neural Network
• Masked Autoregressive Flow for Density Estimation
• NF-ULA: Langevin Monte Carlo with Normalizing Flow Prior for Imaging Inverse Problems
• The frontier of simulation-based inference

This page was built for publication: Simulation-Efficient Double-Flow Neural Posterior Estimation for Agent-Based Models

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7408845)