Introduction to Active Automata Learning from a Practical Perspective

Abstract

Abstract. In this chapter we give an introduction to active learning of Mealy machines, an automata model particularly suited for modeling the behavior of realistic reactive systems. Active learning is characterized by its alternation of an exploration phase and a testing phase. During exploration phases so-called membership queries are used to construct hypothesis models of a system under learning. In testing phases so-called equivalence queries are used to compare respective hypothesis models to the actual system. These two phases are iterated until a valid model of the target system is produced. We will step-wisely elaborate on this simple algorithmic pattern, its underlying correctness arguments, its limitations, and, in particular, ways to overcome apparent hurdles for practical application. This should provide students and outsiders of the field with an intuitive account of the high potential of this challenging research area in particular concerning the control and validation of evolving reactive systems. MotivationInteroperability remains a fundamental challenge when connecting heterogeneous systems [10]. The Connect Integrated Project [35] aims at overcoming the interoperability barrier by synthesizing required Connectors on the fly in five steps [5,21]: (i) extracting knowledge from, (ii) learning about and (iii) reasoning about the interaction behavior of networked systems, as a basis for (iv) synthesizing Connectors [33,34], and subsequently (v) generating and deploying them for immediate use [9].This chapter focuses on the foundations for step (ii), namely on techniques for leveraging and enriching the extracted knowledge by means of experimentation with the targeted components. Central are here advanced techniques for active automata learning [3,38,4,31,50], which are designed for optimally aggregating, and where necessary completing, the observed behavior.Characteristic for active learning automata learning is its iterative alternation between a "testing" phase for completing the transitions relation of the model aggregated from the observed behavior, and an equivalence checking phase, which This work is supported by the European FP 7 project CONNECT (IST 231167).

Combining Model Learning and Model Checking to Analyze TCP Implementations

“…Model learning, or active automata learning [40,6,1], is emerging as a highly effective technique to obtain models of protocol implementations. In fact, all the standard violations reported in [17,37,13,42] have been discovered (or reconfirmed) with the help of model learning.…”

“…Nevertheless, through application of conformance testing algorithms [26], we may increase confidence in the correctness of the learned models. In many recent studies, state-of-the-art tools such as LearnLib [40] routinely succeeded to learn correct models efficiently. In the absence of a tractable white-box model of a protocol implementation, a learned model is often an excellent alternative that may be obtained at relatively low cost.…”

Inferring Semantic Interfaces of Data Structures

“…In classical Mealy machine learning, words are recognized as leading to the same state if they have the same residual semantics [16], i.e., the same output for all suffixes. This requirement has to be loosened slightly, since we have to abstract from concrete data values while still respecting (in-)equalities between data values.…”

LearnLib Tutorial: From Finite Automata to Register Interface Programs

“…However, the testbased interaction introduces a number of challenges when using active automata learning to infer models of real word systems, which have been summarized in [21]:…”