Integrating Gaussian Process and C-Mixup for Regression
Xinxin Li, Jie Zhou · 2025
This paper presents a novel approach that integrates Gaussian Process Regression (GPR) with C-Mixup, aiming to explore the synergistic potential of these two techniques in regression tasks. The GPR offers a flexible and powerful probabilistic framework for regression tasks, while C-Mixup is a data augmentation method that can enhance the generalization ability of models. By integrating these two techniques, we construct a model that performs better than the standard GPR model in regression performance. This model innovatively integrates the powerful function fitting ability of GPR and the unique advantages of C-Mixup in expanding data diversity. The experimental results indicate that under certain conditions, the proposed method outperforms the standard GPR with respect to Mean Squared Error (MSE) and the area of confidence region. Moreover, its sensitivity to the hyper-parameters is lower than that of the standard GPR. The experimental results validate the effectiveness of our proposed method.