Dynestyx: A Probabilistic Programming Library for Dynamical Systems

2026-04-27 · Source: stat.ML updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Mathematics & Computational Sciences · Depth: Advanced, long

Summary

Dynestyx is a new probabilistic programming library designed to provide first-class support for state-space models (SSMs) within dynamical systems. Extending the NumPyro PPL, dynestyx offers a unified interface for specifying arbitrary priors for discrete-time or continuous-time dynamical systems, performing inference over mixed-effect data, and making state and parameter estimates with principled uncertainty quantification. It addresses the historical difficulty of incorporating dynamical systems into modern PPLs by orthogonalizing model specification, marginal likelihood estimation, and parameter inference. The library integrates with existing software such as dynamax, cd-dynamax, and cuthbert, enabling a wide range of advanced state estimation algorithms like Kalman filtering, extended Kalman filtering, and particle filtering. Dynestyx models are compatible with generic JAX routines for parameter inference, including variational inference and HMC/NUTS, and introduce previously unimplemented combinations of methods.

Key takeaway

For Machine Learning Engineers or Data Scientists implementing Bayesian inference for dynamical systems, dynestyx simplifies complex workflows. You can now specify models and interchange advanced inference methods, including various Kalman and particle filters, through a unified interface. This enables rapid development and systematic comparison of algorithms, allowing you to achieve principled uncertainty quantification and explore novel method combinations for your time-series analysis.

Key insights

Dynestyx unifies state-space model specification and Bayesian inference, making advanced methods accessible in probabilistic programming.

Principles

• Separate model specification from inference methods.
• Exploit rich, recursive structure of SSMs for inference.
• Combine diverse ML estimators and parameter inference.

Method

Specify a model as a Python function defining joint priors, packaging them into a DynamicalModel, and sampling with dynestyx.sample(...). Inference then uses user-specified effect handlers for approximate marginal likelihood computation.

In practice

• Use dynestyx for Bayesian inference in dynamical systems.
• Explore various Kalman and particle filtering algorithms.
• Apply generic JAX routines for parameter estimation.

Topics

• Probabilistic Programming
• State-Space Models
• Bayesian Inference
• NumPyro
• Kalman Filtering
• Particle Filtering

Code references

• BasisResearch/dynestyx
• state-space-models/cuthbert
• hd-UQ/cd_dynamax

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Variance Computation for Weighted Model Counting with Knowledge Compilation Approach — Computing WMC variance is tractable for structured d-DNNFs but intractable for broader DNNF variants.
• Geometric Dictionary Learning of Dynamical Systems with Optimal Transport — DOODL learns shared spectral manifolds of dynamical systems for robust, data-efficient operator estimation and system identification.
• My SciPy ODE Solver Was Killing My Bayesian Inference: A Cosmologist’s Honest Account of Discovering Diffrax — JAX-based `diffrax` ODE solvers significantly accelerate Bayesian inference by enabling JIT compilation, autodiff, and vectorization.
• Closing the Approximation Gap in Simulation-free Latent SDEs — Simulation-free VI for latent SDEs can restrict posterior path laws, degrading inference; Helmholtz-SDE resolves this.
• Deep Learning Surrogates for Emulating Stochastic Climate Tipping Dynamics — A dynamics-informed Temporal Fusion Transformer provides a 465x faster, differentiable surrogate for forecasting stochastic climate tipping events with high fidelity.
• Tempered Sequential Monte Carlo for Trajectory and Policy Optimization with Differentiable Dynamics — TSMC optimizes trajectories and policies by framing controller design as inference, using tempered sampling and gradient-based rejuvenation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.