Aether Catalog definitions — Logic_StoneDualityNeuralNetworks
DefinitionLogic_StoneDualityNeuralNetworksaether-cataloglogic
Definition bundle for the Aether Catalog module Logic.StoneDualityNeuralNetworks, transplanted by skeleton subtraction; supplies the types and constants the catalog's theorems import.
Definition code
-- Def bundle generated from Logic/StoneDualityNeuralNetworks.lean by skeleton subtraction
import Mathlib
/-!
# Stone duality and neural classifiers: precise positive results and counterexamples
A classifier induces a Boolean algebra of observable predicates, but several stronger
claims in the proposed framing fail. This file isolates the failures without assuming
any particular implementation of neural networks.
-/
namespace StoneDualityNeuralNetworks
/-- The strict-threshold binary classifier on the real line. -/
noncomputable def thresholdClassifier (x : ℝ) : Bool := decide (0 < x)
/-- Its positive decision region. -/
def positiveRegion : Set ℝ := {x | thresholdClassifier x = true}
/-- Its nonpositive decision region. -/
def nonpositiveRegion : Set ℝ := {x | thresholdClassifier x = false}
/-- Activation vector of a list of scalar threshold neurons. -/
noncomputable def activationVector {k : ℕ} (bias : Fin k → ℝ) (x : ℝ) : Fin k → Bool :=
fun i => decide (bias i < x)
/-- A hypothesis class is a set of subsets, viewed as positive regions. -/
abbrev HypothesisClass (α : Type*) := Set (Set α)
/-- Standard finite-set shattering. -/
def Shatters {α : Type*} (H : HypothesisClass α) (C : Finset α) : Prop :=
∀ L : Finset α, L ⊆ C → ∃ h ∈ H, (C : Set α) ∩ h = L
/-- Pullback of a set along any map; this is the correct elementary mechanism by which
clopens of a finite discrete pattern space yield classifier predicates. -/
def pullback {α β : Type*} (encode : α → β) (s : Set β) : Set α := encode ⁻¹' s
end StoneDualityNeuralNetworks