Concolic testing combines program execution and symbolic analysis to explore the execution paths of a software program. This paper presents the first concolic testing approach for Deep Neural Networks (DNNs). More specifically, we formalise coverage criteria for DNNs that have been studied in the literature, and then develop a coherent method for performing concolic testing to increase test coverage. Our experimental results show the effectiveness of the concolic testing approach in both achieving high coverage and finding adversarial examples.

Abstract. The number of interleavings of a concurrent program makes automatic analysis of such software very hard. Modern multiprocessors' execution models make this problem even harder. Modelling program executions with partial orders rather than interleavings addresses both issues: we obtain an efficient encoding into integer difference logic for bounded model checking that enables first-time formal verification of deployed concurrent systems code. We implemented the encoding in the CBMC tool and present experiments over a wide range of memory models, including SC, Intel x86 and IBM Power. Our experiments include core parts of PostgreSQL, the Linux kernel and the Apache HTTP server.

Predicate abstraction is a major method for verification of software. However, the generation of the abstract Boolean program from the set of predicates and the original program suffers from an exponential number of theorem prover calls as well as from soundness issues. This paper presents a novel technique that uses an efficient SAT solver for generating the abstract transition relations of ANSI-C programs. The SAT-based approach computes a more precise and safe abstraction compared to existing predicate abstraction techniques.

CBMC implements bit-precise bounded model checking for C programs and has been developed and maintained for more than ten years. CBMC verifies the absence of violated assertions under a given loop unwinding bound. Other properties, such as SV-COMP's ERROR labels or memory safety properties are reduced to assertions via automated instrumentation. Only recently support for efficiently checking concurrent programs, including support for weak memory models, has been added. Thus, CBMC is now capable of finding counterexamples in all of SV-COMP's categories. As back end, the competition submission of CBMC uses MiniSat 2.2.0.

Abstract-The software in an electronic system is often the greatest concern with respect to quality and design flaws. Formal verification tools can provide a guarantee that a design is free of specific flaws. We survey algorithms that perform automatic, static analysis of software to detect programming errors or prove their absence. The three techniques we consider are static analysis with abstract domains, model checking, and bounded model checking. We provide a short tutorial on the these techniques, highlighting their differences when applied to practical problems. We also survey the tools that are available implementing these techniques, and describe their merits and shortcomings.