A DPLL(T) Framework for Verifying Deep Neural Networks

07/17/2023
by   Hai Duong, et al.
0

Deep Neural Networks (DNNs) have emerged as an effective approach to tackling real-world problems. However, like human-written software, automatically-generated DNNs can have bugs and be attacked. This thus attracts many recent interests in developing effective and scalable DNN verification techniques and tools. In this work, we introduce a NeuralSAT, a new constraint solving approach to DNN verification. The design of NeuralSAT follows the DPLL(T) algorithm used modern SMT solving, which includes (conflict) clause learning, abstraction, and theory solving, and thus NeuralSAT can be considered as an SMT framework for DNNs. Preliminary results show that the NeuralSAT prototype is competitive to the state-of-the-art. We hope, with proper optimization and engineering, NeuralSAT will carry the power and success of modern SAT/SMT solvers to DNN verification. NeuralSAT is avaliable from: https://github.com/dynaroars/neuralsat

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset