Abstract This paper serves as a comprehensive system description of version 2.0 of the Marabou framework for formal analysis of neural networks. We discuss the tool’s architectural design and highlight the major features and components introduced since its initial release.

Abstract Inspired by sum-of-infeasibilities methods in convex optimization, we propose a novel procedure for analyzing verification queries on neural networks with piecewise-linear activation functions. Given a convex relaxation which over-approximates the non-convex activation functions, we encode the violations of activation functions as a cost function and optimize it with respect to the convex relaxation. The cost function, referred to as the Sum-of-Infeasibilities (SoI), is designed so that its minimum is zero and achieved only if all the activation functions are satisfied. We propose a stochastic procedure, , to efficiently minimize the SoI. An extension to a canonical case-analysis-based complete search procedure can be achieved by replacing the convex procedure executed at each search state with . Extending the complete search with achieves multiple simultaneous goals: 1) it guides the search towards a counter-example; 2) it enables more informed branching decisions; and 3) it creates additional opportunities for bound derivation. An extensive evaluation across different benchmarks and solvers demonstrates the benefit of the proposed techniques. In particular, we demonstrate that SoI significantly improves the performance of an existing complete search procedure. Moreover, the SoI-based implementation outperforms other state-of-the-art complete verifiers. We also show that our technique can efficiently improve upon the perturbation bound derived by a recent adversarial attack algorithm.

Abstract The inference of program invariants over machine arithmetic, commonly called bit-vector arithmetic, is an important problem in verification. Techniques that have been successful for unbounded arithmetic, in particular Craig interpolation, have turned out to be difficult to generalise to machine arithmetic: existing bit-vector interpolation approaches are based either on eager translation from bit-vectors to unbounded arithmetic, resulting in complicated constraints that are hard to solve and interpolate, or on bit-blasting to propositional logic, in the process losing all arithmetic structure. We present a new approach to bit-vector interpolation, as well as bit-vector quantifier elimination (QE), that works by lazy translation of bit-vector constraints to unbounded arithmetic. Laziness enables us to fully utilise the information available during proof search (implied by decisions and propagation) in the encoding, and this way produce constraints that can be handled relatively easily by existing interpolation and QE procedures for Presburger arithmetic. The lazy encoding is complemented with a set of native proof rules for bit-vector equations and non-linear (polynomial) constraints, this way minimising the number of cases a solver has to consider. We also incorporate a method for handling concatenations and extractions of bit-vector efficiently.

Inspired by recent successes of parallel techniques for solving Boolean satisfiability, we investigate a set of strategies and heuristics to leverage parallelism and improve the scalability of neural network verification. We present a general description of the Split-and-Conquer partitioning algorithm, implemented within the Marabou framework, and discuss its parameters and heuristic choices. In particular, we explore two novel partitioning strategies, that partition the input space or the phases of the neuron activations, respectively. We introduce a branching heuristic and a direction heuristic that are based on the notion of polarity. We also introduce a highly parallelizable pre-processing algorithm for simplifying neural network verification problems. An extensive experimental evaluation shows the benefit of these techniques on both existing and new benchmarks. A preliminary experiment ultra-scaling our algorithm using a large distributed cloud - based platform also shows promising results.

Deep neural networks are revolutionizing the way complex systems are designed. Consequently, there is a pressing need for tools and techniques for network analysis and certification. To help in addressing that need, we present Marabou, a framework for verifying deep neural networks. Marabou is an SMT-based tool that can answer queries about a network’s properties by transforming these queries into constraint satisfaction problems. It can accommodate networks with different activation functions and topologies, and it performs high-level reasoning on the network that can curtail the search space and improve performance. It also supports parallel execution to further enhance scalability. Marabou accepts multiple input formats, including protocol buffer files generated by the popular TensorFlow framework for neural networks. We describe the system architecture and main components, evaluate the technique and discuss ongoing work.

The inference of program invariants over machine arithmetic, commonly called bit-vector arithmetic, is an important problem in verification. Techniques that have been successful for unbounded arithmetic, in particular Craig interpolation, have turned out to be difficult to generalise to machine arithmetic: existing bit-vector interpolation approaches are based either on eager translation from bit-vectors to unbounded arithmetic, resulting in complicated constraints that are hard to solve and interpolate, or on bit-blasting to propositional logic, in the process losing all arithmetic structure. We present a new approach to bit-vector interpolation, as well as bit-vector quantifier elimination (QE), that works by lazy translation of bit-vector constraints to unbounded arithmetic. Laziness enables us to fully utilise the information available during proof search (implied by decisions and propagation) in the encoding, and this way produce constraints that can be handled relatively easily by existing interpolation and QE procedures for Presburger arithmetic. The lazy encoding is complemented with a set of native proof rules for bit-vector equations and non-linear (polynomial) constraints, this way minimising the number of cases a solver has to consider.

We consider the problem of automatically and efficiently computing models of constraints, in the presence of complex background theories such as floating-point arithmetic. Constructing models, or proving that a constraint is unsatisfiable, has various applications, for instance for automatic generation of test inputs. It is well-known that a naïve encoding of constraints into simpler theories (for instance, bit-vectors or propositional logic) often leads to a drastic increase in size, or that it is unsatisfactory in terms of the resulting space and runtime demands. We define a framework for systematic application of approximations in order to improve performance. Our method is more general than previous techniques in the sense that approximations that are neither under- nor over-approximations can be used, and it shows promising performance on practically relevant benchmark problems.