NeuPI: A Library for Neural Probabilistic Inference
Authors: S. Arya, T. Rahman, V. Gogate
NeuPI is a PyTorch-based library for solving inference tasks in Probabilistic Models (PMs) using neural network surrogates. It provides a modular, self-supervised framework where the probabilistic model itself supplies the supervisory signal—eliminating the need for labeled training data. By learning to approximate inference directly from model structure, NeuPI reduces reasoning time from minutes to microseconds while maintaining high fidelity to exact inference results.
Key features:
- Self-Supervised Training: Trains neural surrogates using only the PM definition, requiring no labeled data.
- Advanced Inference: Employs the ITSELF (Inference-Time Self-Supervised Learning) engine for test-time adaptation, significantly improving inference accuracy.
- Extensible Design: Built-in factory system for registering custom components
Supported Tasks: Most Probable Explanation (MPE), Marginal MAP, Constrained MPE
Supported Models: Probabilistic Graphical Models, Probabilistic Circuits, Neural Autoregressive Models
The library implements award-winning methods from top-tier AI/ML conferences providing both fast single-pass inference and advanced test-time adaptation capabilities for scalable probabilistic reasoning.