We investigate exact learning of formulas in finite-variable logics over positively- and negatively-labeled finite structures. We propose a general automata-theoretic solution, establish decidability, and give algorithms for synthesis. We also study variants of the learning problem for learning queries and terms, and we provide algorithms for realizability and synthesis along with upper and lower bounds. We study the problem also along the dimension of logics, establishing positive results for first-order logic with least fixed points and several other logics.
翻译:我们研究精确地学习在有限可变逻辑中的公式,而不是正面和负面标签的有限结构。我们提出一个通用的自动化理论解决方案,确定可变性,并为合成提供算法。我们还研究学习查询和术语学习学习问题的变体,并提供可变和合成的算法,以及上下界和上下界。我们还在逻辑层面研究这一问题,用最低固定点和若干其他逻辑为一阶逻辑取得积极结果。