Challenging SMT solvers to verify neural networks

@article{Pulina2012ChallengingSS,
  title={Challenging SMT solvers to verify neural networks},
  author={Luca Pulina and Armando Tacchella},
  journal={AI Commun.},
  year={2012},
  volume={25},
  pages={117-135}
}
In recent years, Satisfiability Modulo Theory (SMT) solvers are becoming increasingly popular in the Computer Aided Verification and Reasoning community. Used natively or as back-engines, they are accumulating a record of success stories and, as witnessed by the annual SMT competition, their performances and capacity are also increasing steadily. Introduced in previous contributions of ours, a new application domain providing an outstanding challenge for SMT solvers is represented by… Expand
Formal Verification of Piece-Wise Linear Feed-Forward Neural Networks
TLDR
An approach for the verification of feed-forward neural networks in which all nodes have a piece-wise linear activation function and infers additional node phases for the non-linear nodes in the network from partial node phase assignments, similar to unit propagation in classical SAT solving. Expand
Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks
TLDR
Results show that the novel, scalable, and efficient technique presented can successfully prove properties of networks that are an order of magnitude larger than the largest networks verified using existing methods. Expand
Verification of Neural Nets : Towards Deterministic Guarantees
The success of neural networks in different classification and regression tasks encouraged different researchers and companies to incorporate them in safetycritical systems such as autonomousExpand
Advances in verification of ReLU neural networks
TLDR
A solver for verification of ReLU neural networks which combines mixed integer programming with specialized branching and approximation techniques is implemented and able to solve the verification problem for instances which do not have independent bounds for each input neuron. Expand
Towards Scalable Verification of RL-Driven Systems
TLDR
This work presents the whiRL 2.0 tool, which implements a new approach for verifying complex properties of interest for DRL systems, and proposes techniques for performing k-induction and automated invariant inference on such systems. Expand
Verifying Quantized Neural Networks using SMT-Based Model Checking
TLDR
A symbolic verification framework using incremental model checking (IMC) and satisfiability modulo theories (SMT) to check for vulnerabilities in ANNs is developed and several ANN-related optimizations for IMC are proposed, including invariant inference via interval analysis and the discretization of non-linear activation functions. Expand
Parallel SMT Solving and Concurrent Symbolic Execution
TLDR
This paper designs, prototypical implementation, and evaluation of a linearly scalable SMT solver cluster and an extension of the KLEE symbolic execution engine, offering concurrent execution and asynchronous constraint solving, and shows that, depending on the characteristics of the program being analyzed, KLEe's performance is improved by up to 7.6x with the help of the approach. Expand
Towards Scalable Verification of Deep Reinforcement Learning
TLDR
This work presents the whiRL 2.0 tool, which implements a new approach for verifying complex properties of interest for DRL systems, and proposes techniques for performing k-induction and semi-automated invariant inference on such systems. Expand
Automated Verification of Neural Networks: Advances, Challenges and Perspectives
TLDR
A primer of such techniques and a comprehensive categorization of existing approaches for the automated verification of neural networks is proposed and a discussion about current limitations and directions for future investigation is provided to foster research on this topic at the crossroads of Machine Learning and Automated Reasoning. Expand
Reluplex: a calculus for reasoning about deep neural networks
TLDR
A novel, scalable, and efficient technique based on the simplex method, extended to handle the non-convex Rectified Linear Unit (ReLU) activation function, which is a crucial ingredient in many modern neural networks. Expand
...
1
2
3
4
5
...

References

SHOWING 1-10 OF 44 REFERENCES
Checking Safety of Neural Networks with SMT Solvers: A Comparative Evaluation
In this paper we evaluate state-of-the-art SMT solvers on encodings of verification problems involving Multi-Layer Perceptrons (MLPs), a widely used type of neural network. Verification is a keyExpand
An Abstraction-Refinement Approach to Verification of Artificial Neural Networks
TLDR
A solution to verify their safety using abstractions to Boolean combinations of linear arithmetic constraints, and it is shown that whenever the abstract MLP is declared to be safe, the same holds for the concrete one. Expand
Extending CREST with Multiple SMT Solvers and Real Arithmetic
TLDR
An approach to overcome the limitation of unique solver’s ability by using multiple SMT solvers and combining their results to get the best solution is proposed and a method of reasoning real arithmetic for symbolic testing is proposed. Expand
SMT-COMP: Satisfiability Modulo Theories Competition
TLDR
Of particular recent interest are solvers for Satisfiability Modulo Theories (SMT), which decide logical satisfiability (or dually, validity) with respect to a background theory in classical first-order logic with equality. Expand
Bounded model checking of software using SMT solvers instead of SAT solvers
TLDR
This paper proposes a generalisation of the cbmc approach on the basis of an encoding into richer (but still decidable) theories than propositional logic, and builds a prototype implementation of the technique that uses a satisfiability modulo theories solver to solve the resulting formulae. Expand
A self-adaptive multi-engine solver for quantified Boolean formulas
TLDR
The rewarding results obtained in the experiments show that the solver AQME—Adaptive QBF Multi-Engine—can be more robust and efficient than state-of-the-art single-engine solvers, even when it is confronted with previously uncharted formulas and competitors. Expand
Z3: An Efficient SMT Solver
TLDR
Z3 is a new and efficient SMT Solver freely available from Microsoft Research that is used in various software verification and analysis applications. Expand
A SAT Based Approach for Solving Formulas over Boolean and Linear Mathematical Propositions
TLDR
A DPLL-based SAT procedure, for dealing efficiently with the propositional component of the problem, and a tight integration, within the DPLL architecture, of a set of mathematical deciders for theories of increasing expressive power are presented. Expand
Efficient Solving of Large Non-linear Arithmetic Constraint Systems with Complex Boolean Structure
TLDR
This work provides a tight integration of recent SAT solving techniques with interval-based arithmetic constraint solving, able to handle large constraint systems with extremely complex Boolean structure, involving Boolean combinations of multiple thousand arithmetic constraints over some thousands of variables. Expand
Automated Configuration of Mixed Integer Programming Solvers
TLDR
This work studies the application of an automated algorithm configuration procedure to different MIP solvers, instance types and optimization objectives, and shows that this fully-automated process yields substantial improvements to the performance of three MIPsolvers: Cplex, Gurobi, and lpsolve. Expand
...
1
2
3
4
5
...