Linear-Time Temporal Logic in AI-Based Computational Tree Models

Studies in Logic, Vol. 19, No. 3 (2026): 59–91                                  PII: 1674-3202(2026)-03-0059-33

Chen Liu

Abstract. The integration of the Linear Temporal Logic (LTL) in AI-based computational tree models is an effective form of formal verification of Artificial Intelligence systems. But classical model checking methods do not scale because of the state-space explosion. In this paper, we propose a novel data driven methodology backed by deep learning for efficient approximation of LTL model checking. The Temporal Convolutional Graph Network (TCGN) is a framework for deep learning that integrates graph neural networks with temporal logic in order to verify the behaviour of complex systems. We conduct experiments on synthetic and real-world datasets, showing that TCGN achieves an accuracy of 98% and outperforms traditional methods like NuSMV both in effectiveness instantaneous verification and efficiency. The TCGN can also achieve real time large scale AI system verification due to its strong robustness under adversarial perturbation.