ONNX Probabilistic Programming Working Group – Collaboration with the BayesFlow Community

Announcements
1

Hi everyone,

I wanted to share that a new ONNX Working Group focused on Probabilistic Programming has recently been formed and invite participation from the BayesFlow community.

The goal of this working group is to bring probabilistic modeling and Bayesian inference into the ONNX ecosystem as first-class capabilities, enabling probabilistic models to be represented in a portable intermediate format and executed across frameworks and hardware.

ONNX has traditionally been used to represent neural network models, but the ecosystem is now expanding toward supporting uncertainty-aware models, probabilistic inference, and hybrid deep learning–Bayesian workflows.

Areas we are exploring

The working group is currently exploring several areas relevant to probabilistic computation:

  • Probability distributions and log-probability operators

  • Bijectors and constrained parameter transformations

  • Reproducible stateless RNG semantics for scalable inference

  • Special mathematical functions used in probabilistic models

  • Inference algorithms including Laplace, Pathfinder, INLA, HMC, NUTS, and SMC

  • Export pathways for probabilistic programming frameworks

Frameworks we are looking to support

We are aiming to support a broad range of probabilistic programming and inference frameworks, including:

  • Stan

  • PyMC

  • Pyro

  • NumPyro

  • TensorFlow Probability

  • JAX-based probabilistic systems

  • BayesFlow

  • Julia probabilistic programming frameworks (Turing.jl, RxInfer.jl)

  • INLA-based approaches

Why input from the BayesFlow community matters

BayesFlow represents one of the leading frameworks for simulation-based inference and amortized Bayesian inference, particularly for complex models where traditional likelihood-based methods are difficult to apply.

As the working group explores support for probabilistic inference workflows in ONNX, input from the BayesFlow community will be especially valuable in areas such as:

  • Amortized inference architectures

  • Neural likelihoods and posterior networks

  • Simulation-based workflows

  • Hybrid deep learning and probabilistic inference pipelines

These perspectives will help ensure that ONNX support for probabilistic models can accommodate modern simulation-based inference methods alongside traditional Bayesian inference algorithms.

Getting involved

If you’re interested in participating, contributing ideas, or providing feedback from the BayesFlow perspective, feel free to reach out to:

You are also welcome to attend the working group meetings:

:spiral_calendar: Fridays @ 12 PM EST, every two weeks

Working group repository:

https://github.com/onnx/working-groups/tree/main/probabilistic-programming

We would love to hear perspectives from the BayesFlow community as we work toward building a portable ecosystem for probabilistic modeling and inference.

1 Like