Interpretable Fuzzy Inference Traces for Regulatory Compliance: Lukasiewicz Semantics in Neuralized Markov Logic Networks
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This paper introduces a fuzzy inference framework grounded in Lukasiewicz semantics for interpretable reasoning in high-risk artificial intelligence (AI) systems. The framework, realised through LIMEN-AI (Lukasiewicz Interpretable Markov Engine for Neuralized AI), extends Markov Logic Networks by replacing Boolean satisfiability with piecewise-linear Lukasiewicz t-norms and implications, yielding differentiable energy manifolds whose gradients guide both learning and inference. Every inference step produces a structured, human-readable trace that decomposes conclusions into weighted rule activations and graded satisfaction degrees, providing a level of transparency that satisfies the requirements of the European Union AI Act (Articles 13–15). Unlike post-hoc explanation methods, this interpretability is intrinsic to the fuzzy logic formulation itself. Through systematic empirical validation on legal document analysis, we demonstrate: (1) zero-shot schema adaptation across regulatory domains, (2) knowledge base evolution through inductive learning (0 to 18 predicates without catastrophic forgetting), (3) document processing scalability (220 words/second on regulatory text), and (4) human override mechanisms with localised intervention effects. These results confirm that Lukasiewicz fuzzy semantics can serve as a principled foundation for neuro-symbolic systems requiring both differentiability and auditability. The open-source implementation (v0.2.5 on PyPI) enables community validation and independent verification.
Fields of Study & Subjects
Explainable Artificial Intelligence (XAI)
Fuzzy Logic and Control Systems
Artificial Intelligence in Law
Publication details
- Journal / source
- Research Square
- Publication date
- 31 March 2026
- Language
- EN
- Document type
- Preprint