Combining Neural Networks with Gaussian Processes for Robust and Interpretable Prediction
Amalia Macali (TU Delft - Group Eskue)
Georgios Soimoiris (TU Delft - Group Yaghoubi Nasrabadi)
Nathan Eskue (TU Delft - Group Eskue)
Vahid Yaghoubi (TU Delft - Group Yaghoubi Nasrabadi)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
This paper presents a hybrid model that combines Artificial Neural Networks (ANN) and Gaussian Processes (GP). The goal is to achieve high prediction accuracy while quantifying uncertainty. The proposed structure is a simple ANN used as the trend of the GP, particularly emphasizing the joint training of the two models. The ANN+GP exploits the ability of the ANN to capture complex, non-linear relationships in the data. At the same time, the GP provides an approach to uncertainty estimation, thus improving the accuracy of the predictions. This paper emphasizes the importance of concurrent training, which can improve the accuracy of the prediction model. The algorithm is suitable for any application where both accurate, robust predictions and uncertainty estimates are critical to enhance the interpretability of the model. The proposed method has been successfully applied to the frequency response functions of a simple structure.