CAREER: Decision Procedures for High-Assurance, AI-Controlled, Cyber-Physical Systems

职业：高可信度、人工智能控制、网络物理系统的决策程序

• 批准号: 2002405
• 负责人: Yasser Shoukry
• 金额: $ 49.69万
• 依托单位: University of California-Irvine
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2025-04-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2002405&HistoricalAwards=false
• 关键词: CAREER; Decision; Procedures; Assurance; AI

This project explores new mathematical techniques that provide a scientific basis to understand the fundamental properties of Cyber-Physical Systems (CPS) controlled by Artificial Intelligence (AI) and guide their design. From simple logical constructs to complex deep neural network models, AI agents are increasingly controlling physical/mechanical systems. Self-driving cars, drones, and smart cities are just examples of AI-controlled CPS. However, regardless of the explosion in the use of AI within a multitude of CPS domains, the safety and reliability of these AI-controlled CPS is still an under-studied problem. This project includes activities integrated with education, so as to explore how learning through counterexamples works for AI, and to help with critical thinking skills for young students.This project investigates a new generation of formal method tools that are capable of simultaneously analyzing the cyber components (including AI-agents) and the physical components of CPS. This new generation of formal methods will be used to analyze the safety and reliability of AI-controlled CPS, characterize the environments for which the system is guaranteed to operate correctly and predict their failure at real-time. In addition, this project will address the problem of how to assign the blame of system failures in AI-controlled CPS. The proposed methods will be evaluated over two flagship testbeds (i) autonomous drones and (ii) self-driving cars. The project engages graduate, undergraduate, and high-school students, and reaches out to the scientific community by providing open source implementations of algorithms.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

该项目探索新的数学技术，为理解人工智能（AI）控制的网络物理系统（CPS）的基本特性提供科学依据，并指导其设计。从简单的逻辑结构到复杂的深度神经网络模型，人工智能代理越来越多地控制物理/机械系统。自动驾驶汽车、无人机和智能城市只是人工智能控制CPS的例子。然而，尽管人工智能在许多CPS领域的使用呈爆炸式增长，但这些人工智能控制的CPS的安全性和可靠性仍然是一个未充分研究的问题。本项目通过与教育相结合的活动，探索通过反例学习如何在人工智能中发挥作用，并帮助年轻学生培养批判性思维能力。本项目研究能够同时分析CPS的网络组件（包括AI代理）和物理组件的新一代形式方法工具。这种新一代的形式化方法将用于分析人工智能控制的CPS的安全性和可靠性，表征系统正确运行的环境，并实时预测其故障。此外，该项目将解决如何在AI控制的CPS中分配系统故障的责任的问题。所提出的方法将在两个旗舰测试平台上进行评估：（i）自动驾驶无人机和（ii）自动驾驶汽车。该项目吸引了研究生、本科生和高中生，并通过提供算法的开源实现来接触科学界。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Formal verification of neural network controlled autonomous systems

• DOI: 10.1145/3302504.3311802
• 发表时间: 2018-10
• 期刊: Proceedings of the 22nd ACM International Conference on Hybrid Systems: Computation and Control
• 作者: Xiaowu Sun;Haitham Khedr;Yasser Shoukry

NNSynth: Neural Network Guided Abstraction-Based Controller Synthesis for Stochastic Systems

NNSynth：神经网络引导的随机系统基于抽象的控制器综合

• DOI: 10.1109/cdc51059.2022.9993158
• 发表时间: 2022
• 期刊: 2022 IEEE 61st Conference on Decision and Control (CDC
• 作者: Sun, Xiaowu;Shoukry, Yasser
• 通讯作者: Shoukry, Yasser

Bounding the Complexity of Formally Verifying Neural Networks: A Geometric Approach

• DOI: 10.1109/cdc45484.2021.9683375
• 发表时间: 2020-12
• 期刊: 2021 60th IEEE Conference on Decision and Control (CDC)
• 作者: James Ferlez;Yasser Shoukry

Two-Level Lattice Neural Network Architectures for Control of Nonlinear Systems

• DOI: 10.1109/cdc42340.2020.9304079
• 发表时间: 2020-04
• 期刊: 2020 59th IEEE Conference on Decision and Control (CDC)
• 作者: James Ferlez;Xiaowu Sun;Yasser Shoukry

AReN: assured ReLU NN architecture for model predictive control of LTI systems

• DOI: 10.1145/3365365.3382213
• 发表时间: 2019-11
• 期刊: Proceedings of the 23rd International Conference on Hybrid Systems: Computation and Control
• 作者: James Ferlez;Yasser Shoukry