Exercise 14.3: Sharper Rates via Entropy Integrals chapter 14 entropy integral Note that f↦1n∑i=1nεf(xi) is a sub-Gaussian process with respect to the metric d(f,g)=‖f−g‖n. Therefore, (1)E[supf∈F1n∑i=1nεf(xi)]≲1n∫0DlogN(u,F,‖⋅‖n)du where (2)F={fθ(x)=⟨ϕ(x),θ⟩:θ∈R3,‖f‖∞≤1} and D=supf,g∈F=‖f−g‖n≤2. Note that θ↦ϕ(X)θ is an isomorphism between R3 and col(ϕ(X)) with the norm ‖ϕ(X)θ‖n=‖f‖n. Hence (3)N(u,F,‖⋅‖n)≤N(u,{θ∈R3:‖⟨ϕ(⋅),θ⟩‖∞≤1},‖ϕ(X)⋅‖n)(4)≤N(u,{θ∈R3:‖ϕ(X)θ‖∞≤1},‖ϕ(X)⋅‖n)(5)≤N(u,{θ∈R3:‖ϕ(X)θ‖n≤1},‖ϕ(X)⋅‖n)(6)≤3log(1+2/u), so the entropy integral is finite and independent of n. « Exercise 14.2: Properties of Local Rademacher Complexity Exercise 14.4: Uniform Laws for Kernel Classes » Published on 9 April 2021. Please enable JavaScript to view the comments powered by Disqus.