bayesflow-org/bayesflow

A Python library for simulation-based inference with deep learning

What it solves

BayesFlow simplifies the process of simulation-based inference (SBI). It allows researchers to perform complex Bayesian analysis—such as estimating parameters from data or comparing different models—without needing to manually derive likelihood functions, which are often unavailable in complex scientific simulations.

How it works

It uses deep learning to "amortize" the inference process. Instead of running a new, slow inference process for every new dataset, you train a neural network to learn the relationship between your simulation parameters and the resulting data. Once trained, this network can provide near-instantaneous estimates. It leverages Keras 3 to support multiple backends, meaning you can run your models using JAX, PyTorch, or TensorFlow.

Who it’s for

It is designed for researchers and data scientists working with complex simulators or parametric models who need to perform Bayesian inference, model comparison, or experimental design. It is particularly useful for those who want to integrate modern generative AI techniques into their scientific workflows.

Highlights

  • Multi-backend support via Keras 3 (JAX, PyTorch, or TensorFlow).
  • Supports advanced generative models, including diffusion and consistency models.
  • Includes built-in diagnostic tools to validate and understand inference results.
  • Provides a high-level API for defining workflows with minimal code.
  • Integrates with the PyMC ecosystem for flexible downstream analysis.

Related

  • Project
    keras-team/keras

    Keras 3 is a multi-backend deep learning framework that allows you to build and train models using JAX, TensorFlow, or PyTorch, preventing framework lock-in while providing high-level usability and state-of-the-art performance.

  • Project
    GPflow/GPflow

    A Python library for building Gaussian process models using TensorFlow and TensorFlow Probability for fast, GPU-accelerated inference.

  • Project
    tensorflow/probability

    A library for probabilistic reasoning and statistical analysis that integrates probabilistic methods with deep networks using TensorFlow or JAX.

  • Project
    tensorly/tensorly

    A Python library for tensor learning that provides tools for tensor decomposition and algebra with seamless support for multiple backends like PyTorch, TensorFlow, and JAX.

  • Project
    LongxingTan/Time-series-prediction

    TFTS is a TensorFlow-based library that provides a simplified, unified interface for building and training deep learning models for time series forecasting, classification, and anomaly detection.