.png){kind=icon}
Return
Gaussian Processes for Learning and Control A TUTORIAL WITH EXAMPLES
.png){kind=icon}
PRE
AI
DOI：10.1109/MCS.2018.2851010.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Gaussian processes (GPs) are data-driven machine learning models that have been widely used in tasks such as regression (modeling the behavior of an unknown system or function based on samples) and classification (for example, in pattern recognition). GP methods applied to control tasks still remain (with only a few notable exceptions) largely unexploited by the general controls community. This tutorial discusses key features of GPs: 1) how they allow for automatic, data-driven feature selection, thereby not requiring practitioners to predefine the number and nature of the features that describe the system being modeled, 2) the way in which GPs offer uncertainty measures over predictions, and 3) how GPs offer a principled way to perform budgeted online inference. This tutorial also discusses a few limitations of the basic GP-based approaches and provides a brief overview of several advanced GP models that overcome such limitations. Finally, the tutorial provides concrete examples demonstrating how GPs can be used for adaptive control and off-policy reinforcement learning and to model value functions in optimal control and planning problems and reward functions in inverse optimal control problems.
Keywords:
ADAPTIVE-CONTROL
FEEDBACK-CONTROL
CONVERGENCE
SYSTEMS
PERSISTENCY
VALIDATION
ALGORITHMS
PARAMETER
NETWORKS
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
I
IF:
6.3
Papers:
1.8K
Citations:
4.7K
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSave
ShareSaveShareSaveShareSave
View more