计算机科学
不可能
反例
自动机
等价(形式语言)
主动学习(机器学习)
Java
理论计算机科学
子程序
人工智能
程序设计语言
数学
离散数学
政治学
法学
作者
Malte Isberner,Bernhard Steffen,Falk Howar
标识
DOI:10.1007/978-3-319-23820-3_25
摘要
Active automata learning is a promising technique to generate formal behavioral models of systems by experimentation. The practical applicability of active learning, however, is often hampered by the impossibility of realizing so-called equivalence queries, which are vital for ensuring progress during learning and finally resulting in correct models. This paper discusses the proposed approach of using monitoring as a means of generating counterexamples, explains in detail why virtually all existing learning algorithms are not suited for this approach, and gives an intuitive account of TTT, an algorithm designed to cope with counterexamples of extreme length. The essential steps and the impact of TTT are illustrated via experimentation with LearnLib, a free, open source Java library for active automata learning.
科研通智能强力驱动
Strongly Powered by AbleSci AI