#gaussian-processes

Question 1: Single choice

mediumAI-checkedNearest Neighbors & Kernel Methods

The RBF kernel $k(x, x') = a^2 \exp\left(-\frac{\lVert x - x' \rVert^2}{2\ell^2}\right)$ makes the correlation between $f(x)$ and $f(x')$ fall steadily as $\lVert x - x' \rVert$ grows. For which data is that assumption most clearly wrong?

Hourly electricity demand over a year, where readings exactly 24 hours apart are strongly related1
Air temperature along a highway, where nearby points are more alike than distant ones2
A smooth growth curve measured at irregular, unevenly spaced times3
Noisy sensor readings of a slowly drifting quantity4

Timed test