NSF Young Investigator

在工程系统中使用自适应和学习策略很重要，原因包括增强对操作条件变化和数据和建模错误的鲁棒性，减少对显式编程的需求，以及随着时间的推移提高性能。然而，尽管这些策略已被广泛使用和研究，但某些基本问题，如“适应的最佳策略是什么？”和“需要多少数据？”在很大程度上仍未得到解答。因此，自适应和学习系统的设计通常是以一种特别的方式完成的，从一个应用程序到下一个应用程序获得的经验很少。此外，所使用的算法可以提供的性能保证很少。本研究的目的是为自适应和学习系统的理论基础做出贡献。一些特定的主题包括分析任意干扰下的估计和自适应控制算法，研究系统识别的基本限制，以及理解先验不确定性和反馈在确定系统性能中的作用。主要目标是获得关于自适应和学习系统的基本能力和局限性的结果，以便对此类系统进行更有原则的设计和分析。

The use of adaptive and learing strategies in engineering systems is important for several reasons including increased robustness against changes in operating conditions and errors in data and modeling, reduced need for explicit programming, and improved performance over time. However, although such strategies have been widely used and studied, certain basic questions such as "What are optimal strategies for adaptation?" and "How much data is required?" remain largely unanswered. Therefore, the design of adaptive and learning systems is generally done in an ad hoc manner with little experience gained from on e application to the next. furthermore, very few performance guarantees can be provided for the algorithms used. The goal of this research is to contribute to the theoretical foundations of adaptive and learning systems. Some particular topics to be pursued include the analysis of estimation and adaptive control algorithms under arbitrary disturbances, the study of fundamental limitations in system identification, and understanding the role of prior uncertainty and feedback in determining system performance. The primary goal is to obtain results on the fundamental capabilities and limitations of adaptive and learning systems in order to allow more principled design and analysis of such systems.