JAIR Releases Three New Papers on Sept 17

The Journal of Artificial Intelligence Research (JAIR) released three research articles on September 17 2026. They are:

1. Abduction in Łukasiewicz Logic with Rational Interval Terms – authors Katsumi Inoue and Daniil Kozhemiachenko. This work builds on decades of research into fuzzy logic, initially proposed by Jan Łukasiewicz around 1920 but only gaining traction with the rise of computer science. It explores abductive reasoning using infinitely‑valued Łukasiewicz fuzzy logic over \(\) and introduces interval literals such as \(p\ge c\) or \(p\le c\) for more nuanced reasoning than traditional Boolean logic. Potential applications exist in medical diagnosis and risk assessment where uncertainty is inherent, though scaling to large datasets remains computationally challenging due to the infinite‑valued nature of Łukasiewicz logic.¹

2. Optimal Clustering with Dependent Costs in Bayesian Networks – authors Paul Pao-Yen Wu, Fabrizio Ruggeri, Kerrie Mengersen. The paper addresses optimal cluster mappings within Bayesian networks, a problem related to feature selection techniques used extensively since Hastie's seminal “Vast Search Spaces” (2003). The authors propose DCMAP, an algorithm that explicitly accounts for dependencies between clusters, offering improvements over existing algorithms. Benchmarks on seagrass DBN configurations show it can find all least‑cost solutions efficiently; further validation against real‑world data sets would be needed before widespread adoption.¹

3. Certified Neural Networks: from Verification to Synthesis – authors Matteo Zavatteri, Davide Bresolin, Nicolò Navarin. This research tackles formal verification’s limitation by combining deep learning with rigorous proofs through a Counter Example Guided Inductive Synthesis loop. It automatically builds certified neural networks guaranteed to respect required properties expressed as logical constraints. Synthesizing certified controllers could revolutionize robotics applications such as autonomous vehicles or surgical robots while also potentially opening new avenues for adversarial attacks on these synthesized models.¹