Jeremy Gillen comments on Neural Tangent Kernel Distillation

[Jeremy Gillen]

The eigenfunctions we are calculating are solutions to:

$\lambda \varphi \left(x^{\prime }\right)={\int }_{x\sim D}k(x^{\prime },x)\varphi \left(x\right)dx$

Where $D$ is the data distribution, $\lambda$ is an eigenvalue and $\varphi \left(x\right)$ is an eigenfunction.

So the eigenfunction is a label function with input $x$, a datapoint. The discrete approximation to it is a label vector, which I called $y$ above.