Introduction to BayesFlow
Stefan Radev is an assistant professor at Rensselaer Polytechnic Institute and the lead maintainer of the software package BayesFlow.
Simulation-Based Inference With Neural Networks: From Theory to Practice
Modern Bayesian inference combines computational techniques for estimating, validating, and drawing conclusions from probabilistic models within principled statistical workflows. Recently, a suite of simulation-based inference (SBI) methods has emerged as a key catalyst for scaling these workflows to complex Bayesian models and large data sets.
This workshop will introduce participants to the core theory and concepts of SBI with neural networks, using the BayesFlow software for amortized Bayesian inference.
BayesFlow supports multiple backends, including PyTorch, TensorFlow, and JAX, and offers flexible generative networks for sampling, full customization, high-level interfaces, and tools for hyperparameter tuning, design optimization, and estimation from long time series.
Attendees will receive a comprehensive overview of the SBI landscape and follow a hands-on path from model building to inference and model-based prediction. The workshop will cover a variety of data types, practical applications, and diagnostic tools. Participants will gain insights into the strengths and limitations of specific SBI techniques, along with the potential synergies between different approaches and existing software libraries.