Falk Howar

Introduction to Active Automata Learning from a Practical Perspective

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.

Towards an architecture for runtime interoperability

Interoperability remains a fundamental challenge when connecting heterogeneous systems which encounter and spontaneously communicate with one another in pervasive computing environments. This challenge is exasperated by the highly heterogeneous technologies employed by each of the interacting parties, i.e., in terms of hardware, operating system, middleware protocols, and application protocols. This paper introduces CONNECT, a software framework which aims to resolve this interoperability challenge in a fundamentally different way. CONNECT dynamically discovers information about the running systems, uses learning to build a richer view of a systems behaviour and then uses synthesis techniques to generate a connector to achieve interoperability between heterogeneous systems. Here, we introduce the key elements of CONNECT and describe its application to a distributed marketplace application involving heterogeneous technologies.

Reusing System States by Active Learning Algorithms

In this paper we present a practical optimization to active automata learning that reduces the average execution time per query as well as the number of actual tests to be executed. Key to our optimization are two observations: (1) establishing well-defined initial conditions for a test (reset) is a very expensive operation on real systems, as it usually involves modifications to the persisted state of the system (e.g., a database). (2) In active learning many of the (sequentially) produced queries are extensions of previous queries. We exploit these observations by using the same test run on a real system for multiple “compatible” queries. We maintain a pool of runs on the real system (system states), and execute only suffixes of queries on the real system whenever possible. The optimizations allow us to apply active learning to an industry-scale web-application running on an enterprise platform: the Online Conference Service (OCS) an online service-oriented manuscript submission and review system.

Machine learning for emergent middleware

Highly dynamic and heterogeneous distributed systems are challenging todays middleware technologies. Existing middleware paradigms are unable to deliver on their most central promise, which is offering interoperability. In this paper, we argue for the need to dynamically synthesise distributed system infrastructures according to the current operating environment, thereby generating Emergent Middleware to mediate interactions among heterogeneous networked systems that interact in an ad hoc way. The paper outlines the overall architecture of Enablers underlying Emergent Middleware, and in particular focuses on the key role of learning in supporting such a process, spanning statistical learning to infer the semantics of networked system functions and automata learning to extract the related behaviours of networked systems.

Active Automata Learning in Practice

Active automata learning is slowly becoming a standard tool in the toolbox of the software engineer. As systems become ever more complex and development becomes more distributed, inferred models of system behavior become an increasingly valuable asset for understanding and analyzing a system’s behavior. Five years ago (in 2011) we have surveyed the then current state of active automata learning research and applications of active automata learning in practice. We predicted four major topics to be addressed in the then near future: efficiency, expressivity of models, bridging the semantic gap between formal languages and analyzed components, and solutions to the inherent problem of incompleteness of active learning in black-box scenarios. In this paper we review the progress that has been made over the past five years, assess the status of active automata learning techniques with respect to applications in the field of software engineering, and present an updated agenda for future research.

Model-Based Testing Without Models: The TodoMVC Case Study.

Web applications define the interface to many of the businesses and services that we interact with and use on a daily basis. The technology stack enabling these applications is constantly changing and applications are accessed from a plethora of different devices. Automated testing of the behavior of applications is a promising strategy for reducing the manual effort that has to be spent on ensuring a consistent user experience across devices. Unfortunately, specifications or models of the desired behavior often do not exist. Model-based testing without models (aka learning-based testing) tries to overcome this hurdle by integrating model learning and model-based testing. In this paper, we sketch the ALEX tool [1, 11] for learning-based testing of web application and demonstrate its operation on benchmarks from the TodoMVC project. Our learning-based conformance analysis reveals that 7 of 27 Todo-apps exhibit behavior that differs from the majority of implementations.