一句话：提出架构保持的 V-polytope 可证明修复方法，仅改参数即可让 DNN 在指定凸多面体上满足规格，并实现工具 APRNN。
深度神经网络（DNN）日益成为软件的重要组件，但其错误行为可能带来严重后果。本文研究保持架构的 DNN 可证明修复：给定由顶点表示的 V-polytope，修复须保证 DNN 在该无限点集上满足规格，且只修改参数而不改变网络架构。方法将修复转化为参数更新，允许修改多层，支持含部分线性片段的激活函数以及全连接、卷积、池化和残差层，并保证多项式时间运行。作者实现工具 APRNN，在 MNIST、ImageNet 和 ACAS Xu 上评估，与不保持架构的 PRDNN、REASSURE 相比，在效率、可扩展性和泛化上更优。据称这是首个同时具备上述全部特性的可证明修复方法。
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.
@article{paperbot1945,
title = {Architecture-Preserving Provable Repair of Deep Neural Networks},
author = {Zhe Tao and Stephanie Nawas and Jacqueline Mitchell and Aditya V. Thakur},
journal = {Proceedings of the ACM on Programming Languages},
volume = {7},
number = {PLDI},
year = {2023},
doi = {10.1145/3591238}
}