Neural Tangent Kernel based regression task for small data sets
Regression is a common class of tasks,and two special types of regression models,which means Support Vector Regression(SVR)and Kernel Ridge Regression(KRR)solve the linearly inseparable problem of data in the original space by means of kernel functions.A new Neural Tangent Kernel(NTK)has been proposed to fit the training process of infinitely wide neural networks,and the related studies have shown that NTK is beneficial for handling small datasets.A multi-domain data set is selected to compare the performance of NTK with commonly used kernels in two models,and the robustness of NTK is investigated.The results show that the NTK-SVR model achieves a 2.5％～20％improvement on some data sets.
万方数据
维普
