Read a finite-class generalization bound

Tutor me on https://mlprep.iwase.dev/item/original-ml-finite-class-bound/. If window.mlPrepAgent is available, read attempts for item original-ml-finite-class-bound before tutoring. Inspect my attempt, keep the item ID, and do not reveal the full answer first. After a real attempt, append its record with recorded_by agent, include agent_session_id when available, and read it back.

Problem

A finite hypothesis class has $\lvert\mathcal{H}\rvert=100$ and $n=1000$ independent examples. Using

$\varepsilon=\sqrt{\frac{\log\left(2\lvert\mathcal{H}\rvert/\delta\right)}{2n}}$

with $\delta=0.05$ , compute $\varepsilon$ and explain what the simultaneous bound says.

Reveal answer or reference solution

$\varepsilon=\sqrt{\frac{\log(4000)}{2000}}\approx 0.0644.$

With probability at least $0.95$ , every hypothesis in $\mathcal{H}$ has empirical and population error differing by at most this amount under the bound's assumptions.

Local history

Loading attempts saved in this browser…

Appears in

Theory Practice