Architecture-Preserving Provable Repair of Deep Neural Networks
This addresses the critical issue of ensuring safety and reliability in DNNs used in software systems, offering a novel repair approach that preserves network architecture.
The paper tackles the problem of repairing deep neural networks (DNNs) to ensure correct behavior within specified regions, introducing an architecture-preserving V-polytope provable repair method that modifies only parameters and runs in polynomial time. It shows better efficiency, scalability, and generalization compared to prior methods like PRDNN and REASSURE on datasets including MNIST, ImageNet, and ACAS Xu.
Deep neural networks (DNNs) are becoming increasingly important components of software, and are considered the state-of-the-art solution for a number of problems, such as image recognition. However, DNNs are far from infallible, and incorrect behavior of DNNs can have disastrous real-world consequences. This paper addresses the problem of architecture-preserving V-polytope provable repair of DNNs. A V-polytope defines a convex bounded polytope using its vertex representation. V-polytope provable repair guarantees that the repaired DNN satisfies the given specification on the infinite set of points in the given V-polytope. An architecture-preserving repair only modifies the parameters of the DNN, without modifying its architecture. The repair has the flexibility to modify multiple layers of the DNN, and runs in polynomial time. It supports DNNs with activation functions that have some linear pieces, as well as fully-connected, convolutional, pooling and residual layers. To the best our knowledge, this is the first provable repair approach that has all of these features. We implement our approach in a tool called APRNN. Using MNIST, ImageNet, and ACAS Xu DNNs, we show that it has better efficiency, scalability, and generalization compared to PRDNN and REASSURE, prior provable repair methods that are not architecture preserving.