Probability

When it fits

Use it when a TensorFlow or JAX workflow needs explicit probabilistic model components or inference methods, such as hierarchical modeling, Bayesian clustering or latent-variable dimensionality reduction.

Who it is for

• Machine-learning developers adding uncertainty-aware components to neural networks.
• Data scientists and researchers building Bayesian or other probabilistic models.
• TensorFlow or JAX practitioners who need statistical distributions, transformations or inference algorithms.

Documented uses

• Represent probability distributions and statistics with batch and broadcasting semantics.
• Transform random variables with composable bijectors and combine dependent variables in joint distributions.
• Add uncertainty-aware probabilistic layers to neural networks.
• Run sampling-based or optimization-based inference, including MCMC and variational inference.
• Compute Monte Carlo expectations and use stochastic optimization methods.