Lemma 26.9 (contraction, Kakade–Tewari): for coordinatewise ρ-Lipschitz φ, R(φ ∘ A) ≤ ρ R(A)
ProvedUnderstandingML.contraction_lemmacontraction-lemmalipschitzrademacher-complexity
Lemma 26.9 (Contraction lemma). For each , let be a -Lipschitz function, namely for all we have . For let denote the vector . Let . Then . is nonempty and bounded.
Preamble
import Definitions.Def_UnderstandingML_Rademacher open MeasureTheory open scoped InnerProductSpace
Formal statement
namespace UnderstandingML
/-- **Lemma 26.9 (Contraction lemma)** (p. 381). For each `i ∈ [m]`, let `φᵢ : ℝ → ℝ` be a
`ρ`-Lipschitz function. For `a ∈ ℝ^m` let `φ(a) = (φ₁(a₁), …, φ_m(a_m))` and
`φ ∘ A = {φ(a) : a ∈ A}`. Then `R(φ ∘ A) ≤ ρ R(A)`. `A` is nonempty and bounded. -/
theorem contraction_lemma {m : ℕ} (A : Set (Fin m → ℝ)) (hA : A.Nonempty)
(hb : Bornology.IsBounded A) (ρ : NNReal) (φ : Fin m → ℝ → ℝ)
(hφ : ∀ i, LipschitzWith ρ (φ i)) :
rademacher ((fun a i ↦ φ i (a i)) '' A) ≤ ρ * rademacher A := by sorry
end UnderstandingML
Source
Shalev-Shwartz and Ben-David, Understanding Machine Learning: From Theory to Algorithms, Cambridge University Press 2014, doi:10.1017/CBO9781107298019, §26.1.1 pp. 381-382, Lemma 26.9 with its proof
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.