Two neurons having the same threshold can realize only the two constant activation
ProvedStoneDualityNeuralNetworks.duplicate_neurons_rangeaether-cataloglogic
Two neurons having the same threshold can realize only the two constant activation
patterns, not all four elements of Fin 2 → Bool.
theorem StoneDualityNeuralNetworks.duplicate_neurons_range:
Set.range (activationVector (k := 2) (fun _ => 0)) =
{p | p = (fun _ => false) ∨ p = (fun _ => true)} := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Logic/StoneDualityNeuralNetworks.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Logic/StoneDualityNeuralNetworks.lean import Mathlib import Definitions.Def_Logic_StoneDualityNeuralNetworks /-! # 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. -/ open StoneDualityNeuralNetworks
Formal statement
theorem StoneDualityNeuralNetworks.duplicate_neurons_range:
Set.range (activationVector (k := 2) (fun _ => 0)) =
{p | p = (fun _ => false) ∨ p = (fun _ => true)} := by sorrySource