Robust learning of dynamic systems

动态系统的鲁棒学习

• 批准号: DP190102963
• 负责人: Prof Ian Manchester
• 金额: $ 35.47万
• 依托单位: The University of Sydney
• 依托单位国家: 澳大利亚
• 项目类别: Discovery Projects
• 财政年份: 2019
• 资助国家: 澳大利亚
• 起止时间: 2019-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://dataportal.arc.gov.au/RGS/Web/Grants/DP190102963
• 关键词: Robust; learning; dynamic; systems

Robots and other autonomous machines use models of the real world to predict the result of their actions and make decisions, but existing methods used for machine-learning are unreliable in many cases and can be easily fooled. This project aims to make machine-learning of dynamic system models reliable, accurate, and secure. The outcomes of this project will be new models and algorithms that ensure safety and increase accuracy of models learned from data. This project will benefit robotics, control engineering, infrastructure automation, and other fields that demand the capability to model physical systems from limited data. It will also improve cybersecurity by making learning algorithms resilient to deliberate attacks with false data.

机器人和其他自主机器使用真实的世界的模型来预测其行为的结果并做出决策，但用于机器学习的现有方法在许多情况下是不可靠的，并且很容易被愚弄。该项目旨在使动态系统模型的机器学习可靠，准确和安全。该项目的成果将是新的模型和算法，以确保安全性并提高从数据中学习的模型的准确性。该项目将使机器人、控制工程、基础设施自动化和其他需要从有限数据中建模物理系统的领域受益。它还将通过使学习算法能够抵御虚假数据的蓄意攻击来改善网络安全。