论文标题

深神经切线内核和拉普拉斯内核具有相同的RKHS

Deep Neural Tangent Kernel and Laplace Kernel Have the Same RKHS

论文作者

Chen, Lin, Xu, Sheng

论文摘要

我们证明,当两个内核都仅限于Sphere $ \ Mathbb {s}^{d-1-1} $时,我们的深神经切线内核和Laplace内核的复制核Hilbert空间(RKHS)包括相同的功能。此外,我们证明,具有较小功率的指数功率内核(使内核降低平滑)会导致较大的RKHS,而当它仅限于Sphere $ \ Mathbb {s}^{D-1} $时,并且当它在整个$ \ Mathbb {r}^d $中定义时。

We prove that the reproducing kernel Hilbert spaces (RKHS) of a deep neural tangent kernel and the Laplace kernel include the same set of functions, when both kernels are restricted to the sphere $\mathbb{S}^{d-1}$. Additionally, we prove that the exponential power kernel with a smaller power (making the kernel less smooth) leads to a larger RKHS, when it is restricted to the sphere $\mathbb{S}^{d-1}$ and when it is defined on the entire $\mathbb{R}^d$.

扫码加入交流群

加入微信交流群

微信交流群二维码

扫码加入学术交流群,获取更多资源