Noumenal Labs open-sources hybrid-aif, a JAX active inference library
mjdramstead · x · 2026-11-16
Noumenal Labs' Alex Kiefer has open-sourced hybrid-aif (MIT, v0.20.0), a Python/JAX library for production-grade active inference and probabilistic ML, developed over the past year.
Key features:
- Factor-graph primitives beyond the exponential family: conditional categoricals, Gaussians, Gaussian mixtures, switching linear dynamical systems, user-defined nonlinear/deterministic functions, and amortized neural factors
- Hybrid inference engine: models compiled and made differentiable with JAX, with closed-form conjugate message-passing running alongside gradient- and sample-based inference in the same model
- Flexible generative model declaration: control states and parameters are first-class variables, conditionable on other variables or targets for preferences/constraints
- Compositionality: agents and environments described as nested physical systems via a single System primitive
- Fine-grained objectives: standard expected free energy by default, with information-gain and parameter-novelty terms ablatable per variable
A solar-rover demo accompanies the release, plus model comparison and logging tools.
Related event: Noumenal Labs Open-Sources hybrid-aif, a JAX Active Inference Library(2 posts)→
More from Research
- Big Data Cup 2027 opens 60Hz 3D goalie joint and puck tracking data for hockey analytics — StatsbyLopez · 2026-11-16
- Dominik Grimm named TUM professor of ML for sustainable processes and materials — kmborgwardt · 2026-11-16
- Graph learning researcher Dexiong Chen joins HKUST Guangzhou as assistant professor — kmborgwardt · 2026-11-16
- Gamer: A Generalist VLA Foundation Model Trained on Tens of Thousands of Hours of Keyboard-Mouse Gameplay — pmttyji · 2026-11-16
- AI paper authors swarmed by predatory journal spam that smells like a pay-to-publish scam — letandrewcook · 2026-11-16
- Can we fix AI? Peter Bloem's new essay examines data, copyright and synthetic data paths — pbloemesquire · 2026-11-16