machine-learning - 论文 "Deep learning requires rethinking generalization"

Question

我正在读这篇论文Understanding Deep learning requires rethinking generalization我不明白为什么在第 5 页第 2.2 节“含义、Redemacher 复杂度”下说界限是微不足道的？

Since our randomiyation tests suggest that many neural networks fit the training set with random labels perfectly, we expect that Rad(H)=1 for the corresponding model class H. This, of course, a trivial upper bound on the Rademacher complexity that does not lead to useful generalization bounds in realistic settings.

显然我缺少一些关于 Radmacher 的知识，因为我无法理解他们是如何得出这个结论的。如果有人能向我解释一下，我将非常感激

Answer 1

在论文中，函数 h 的边界为 1，因此 Rademacher 复杂度的边界为 1(您将 n 项相加等于 1，然后除以 n)。