Special issue on learning techniques for compositional reasoning

Abstract

This special issue of Formal Methods in System Design (FMSD) was initiated by Professor Ed Clarke and contains a selection of articles on using learning techniques to automate compositional verification in the assume-guarantee style. Model checking is an automated verification technique that can be used to determine whether a concurrent system satisfies certain properties. It works by systematically exploring all the system states, which is intractable for most systems of realistic size, a limitation known as the “state-explosion problem”. Compositional verification has been identified as a promising approach for alleviating state explosion in model checking. This technique decomposes the verification task for the system into simpler verification problems for the individual components of the system. In checking components individually, assume-guarantee reasoning introduces assumptions, which incorporate knowledge of the contexts in which each component is expected to operate. Assumptions have traditionally been defined manually, which has limited the practical impact of assume-guarantee reasoning. Over the last decade, researchers have focused on the automated generation of assumptions for assume-guarantee reasoning. The papers included in this issue all present solutions to automated assumption generation based on learning. The order in which they appear is chronological, based on the publication of the original conference articles that they extend. The first article, “Learning to Divide and Conquer: Applying the L* Algorithm to Automate Assume-Guarantee Reasoning”, outlines the assume guarantee framework that uses the L* learning algorithm to automatically build assumptions, as originally proposed in [3], and presents extensions with symmetric rules and alphabet refinement. While the first article deals with components described as finite labeled transition systems in the context of