Friends,
The new version preview (pre-release) of BayesFlow is out there, along with updated tutorials and completely revamped codebase for the next generation of amortized Bayesian inference.
You can now use any deep learning framework (TensorFlow, PyTorch, JAX) and run entire amortized workflows in a few lines of code.
All networks have been improved for speed and performance. Some speedups are quite dramatic. Many recent advances in generative AI are implemented with flexible settings (e.g., FlowMatching, ConsistencyModels).
All simpler methods are also available (e.g., various point estimators, such as quantile regression, Gaussian, robust estimators, experts summary stats, combinations).
Happy amortizing!