Learning "best" kernels from data in Gaussian process regression. With application to aerodynamics
Authors:
Jean-Luc Akian,
Luc Bonnet,
Houman Owhadi,
Éric Savin
Abstract:
This paper introduces algorithms to select/design kernels in Gaussian process regression/kriging surrogate modeling techniques. We adopt the setting of kernel method solutions in ad hoc functional spaces, namely Reproducing Kernel Hilbert Spaces (RKHS), to solve the problem of approximating a regular target function given observations of it, i.e. supervised learning. A first class of algorithms is…
▽ More
This paper introduces algorithms to select/design kernels in Gaussian process regression/kriging surrogate modeling techniques. We adopt the setting of kernel method solutions in ad hoc functional spaces, namely Reproducing Kernel Hilbert Spaces (RKHS), to solve the problem of approximating a regular target function given observations of it, i.e. supervised learning. A first class of algorithms is kernel flow, which was introduced in the context of classification in machine learning. It can be seen as a cross-validation procedure whereby a "best" kernel is selected such that the loss of accuracy incurred by removing some part of the dataset (typically half of it) is minimized. A second class of algorithms is called spectral kernel ridge regression, and aims at selecting a "best" kernel such that the norm of the function to be approximated is minimal in the associated RKHS. Within Mercer's theorem framework, we obtain an explicit construction of that "best" kernel in terms of the main features of the target function. Both approaches of learning kernels from data are illustrated by numerical examples on synthetic test functions, and on a classical test case in turbulence modeling validation for transonic flows about a two-dimensional airfoil.
△ Less
Submitted 31 August, 2022; v1 submitted 3 June, 2022;
originally announced June 2022.
Freeze the BCI until the user is ready: a pilot study of a BCI inhibitor
Authors:
Laurent George,
Laurent Bonnet,
Anatole Lécuyer
Abstract:
In this paper we introduce the concept of Brain-Computer Interface (BCI) inhibitor, which is meant to standby the BCI until the user is ready, in order to improve the overall performance and usability of the system. BCI inhibitor can be defined as a system that monitors user's state and inhibits BCI interaction until specific requirements (e.g. brain activity pattern, user attention level) are met…
▽ More
In this paper we introduce the concept of Brain-Computer Interface (BCI) inhibitor, which is meant to standby the BCI until the user is ready, in order to improve the overall performance and usability of the system. BCI inhibitor can be defined as a system that monitors user's state and inhibits BCI interaction until specific requirements (e.g. brain activity pattern, user attention level) are met. In this pilot study, a hybrid BCI is designed and composed of a classic synchronous BCI system based on motor imagery and a BCI inhibitor. The BCI inhibitor initiates the control period of the BCI when requirements in terms of brain activity are reached (i.e. stability in the beta band). Preliminary results with four participants suggest that BCI inhibitor system can improve BCI performance.
△ Less
Submitted 8 November, 2011;
originally announced November 2011.
