UnRAVeL-Ringvorlesung: Sebastian Trimpe: Uncertainty Bounds for Gaussian Process Regression with Applications to Safe Control and Learning
Dienstag, 21.06.2022
UnRAVeL-Ringvorlesung: Sebastian Trimpe: Uncertainty Bounds for Gaussian Process Regression with Applications to Safe Control and Learning
- Dr., Universitätsprofessor Sebastian Trimpe – RWTH Center for Artificial Intelligence
Abstract
Gaussian Process (GP) regression is a popular nonparametric machine learning method that provides uncertainty estimates for its predictions. While GPs are based on Bayesian principles, also frequentist uncertainty bounds are available, which are required for applications in learning-based control or safe reinforcement learning. However, the available uncertainty bounds are typically too conservative to be useful in applications. This often leads practitioners to replacing these bounds by heuristics, thus breaking all theoretical guarantees. To address this problem, we introduce new GP uncertainty bounds that are rigorous, yet practically useful at the same time. In particular, the bounds can be explicitly evaluated and are much less conservative than state of the art results. After an introduction to Gaussian processes, we will discuss these results, and present applications to learning-based control and safe reinforcement learning.