Probabilistic Rule-based Inference and Modeling with Explainable Reasoning
Data-driven AI has made remarkable progress in recent years, but it is not the right tool for every application. Large neural models are expensive to build, may behave erratically, need large amounts of (often sensitive) data, and are hard to analyse with formal methods. Fields such as medical diagnosis, decision support and many scientific applications would benefit from transparent models that can be formally validated and that produce causal explanations understandable by non-experts.
PRIMER develops a framework for probabilistic rule-based modeling: a rule-based formalism for knowledge representation and automated reasoning, extended with probabilities to represent uncertainty. The framework builds on probabilistic term rewriting with a distribution semantics, and will be implemented in open-source tools for specifying models, verifying them, performing inference and visualising explanations.
What is the probability that a given patient has diabetes?
Building on the many analysis and verification techniques available for rule-based formalisms, and extending them to the probabilistic setting.
Models and reasoning mechanisms that are transparent by design, together with the automatic generation of causal explanations.
Exploring the integration of probabilistic rule-based models with machine learning and neural systems (neurosymbolic AI).
| Title | Probabilistic Rule-based Inference and Modeling with Explainable Reasoning |
|---|---|
| Spanish title | Inferencia y modelado basado en reglas probabilísticas con razonamiento explicable |
| Acronym | PRIMER |
| Reference | PID2025-168994OB-I00 |
| Funding | MICIU/AEI/10.13039/501100011033 and ERDF/EU – Proyectos de Generación de Conocimiento 2025 |
| Duration | September 2026 – August 2029 (3 years) |
| Host institution | VRAIN (Valencian Research Institute for Artificial Intelligence), Universitat Politècnica de València |
| Principal investigator | Germán Vidal |