Example 27.1 (as constructed): a set of norm ≤ c in a d-dimensional subspace of ℝ^m has an r-cover of size ≤ (2c√d/r + 1)^d
ProvedUnderstandingML.subspace_coveringcovering-numbersgridsubspace
Example 27.1 (Subspace). Suppose that , let , and assume that lies in a -dimensional subspace of . Then : with an orthonormal basis of the subspace and , the grid is an -cover.
Formally, as the construction gives it: has an -cover of cardinality at most ; the book's count omits the of the grid. The bound is nonnegative, as the book's is. For with and odd the right-hand side would be negative.
Preamble
import Definitions.Def_UnderstandingML_Covering open MeasureTheory
Formal statement
namespace UnderstandingML
/-- **Example 27.1 (Subspace)** (p. 388), as its construction gives it. Suppose `A ⊆ ℝ^m`, let
`c = max_{a ∈ A} ‖a‖`, and assume that `A` lies in a `d`-dimensional subspace of `ℝ^m`. Then
`A` has an `r`-cover of size at most `(2c√d/r + 1)^d`, the grid `{∑ᵢ α'ᵢvᵢ : α'ᵢ ∈ {−c, −c+ε, …, c}}`
with `ε = r/√d` in an orthonormal basis `v₁, …, v_d`. (The book writes `(2c√d/r)^d`, the count
`(2c/ε)^d` omitting the `+1` of the grid; the statement here is what the grid gives.) -/
theorem subspace_covering {m d : ℕ} (A : Set (Fin m → ℝ)) (V : Submodule ℝ (Fin m → ℝ))
(hV : Module.finrank ℝ V = d) (hAV : A ⊆ V) (c : ℝ) (hc0 : 0 ≤ c) (hc : ∀ a ∈ A, eucNorm a ≤ c) (r : ℝ)
(hr : 0 < r) :
∃ A' : Finset (Fin m → ℝ), IsCover r A A' ∧
(A'.card : ℝ) ≤ (2 * c * Real.sqrt d / r + 1) ^ d := 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, §27.1 p. 388, Example 27.1 with its construction
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.