SYSLITE: Syntax-Guided Synthesis of PLTL Formulas from Finite Traces

Abstract

We present an efficient approach to learn past-time linear temporal logic formulas (PLTL) from a set of propositional variables and a sample of finite traces over those variables. The efficiency of our approach can be attributed to a careful encoding of the PLTL formula learning problem as a bit-vector function synthesis problem, and the use of an enhanced Syntax-Guided Synthesis (SyGuS) engine to solve the latter. We implemented our approach in a tool called Syslite and empirically evaluated its efficacy with two case studies. In these case studies, we observe that Syslite on average enjoys a speedup of 44x over current learning approaches for temporal formulas while learning the expected formulas in the vast majority of cases.