标题： 查找ReLU神经网络输出属性的输入表征 显示英文标题

标题： Finding Input Characterizations for Output Properties in ReLU Neural Networks

摘要： 深度神经网络（DNNs）已成为一种强大的机制，并在现实世界的安全关键领域中被越来越多地部署。 尽管取得了广泛的成功，但它们复杂的架构使得证明任何形式的保证变得困难。 确定高层次正确性逻辑概念如何与复杂的低层次网络架构相关联是一个重大挑战。 在这个项目中，我们扩展了和中提出的思想，并引入了一种弥合架构与高层次规范之间差距的方法。 我们的关键见解是，与其直接证明所需的安全部分属性，不如首先证明与神经网络结构密切相关的属性，并利用这些属性来推理安全部分属性。 我们为我们的方法建立了理论基础，并通过各种实验进行了实证评估，通过识别保证输出上某种属性的输入空间更大区域，取得了比现有方法更有希望的结果。 显示英文摘要

摘要： Deep Neural Networks (DNNs) have emerged as a powerful mechanism and are being increasingly deployed in real-world safety-critical domains. Despite the widespread success, their complex architecture makes proving any formal guarantees about them difficult. Identifying how logical notions of high-level correctness relate to the complex low-level network architecture is a significant challenge. In this project, we extend the ideas presented in and introduce a way to bridge the gap between the architecture and the high-level specifications. Our key insight is that instead of directly proving the safety properties that are required, we first prove properties that relate closely to the structure of the neural net and use them to reason about the safety properties. We build theoretical foundations for our approach, and empirically evaluate the performance through various experiments, achieving promising results than the existing approach by identifying a larger region of input space that guarantees a certain property on the output.