Theorem 5.2 — total absolute loss of WMC
ProvedWeightedMajority.Continuous.theorem_5_2Let WMC run on any finite sequence of trials with labels in , using a finite nonempty pool whose predictions lie in . Every initial weight is positive; at each trial, WMC predicts the pool's current weighted mean and updates each weight by any factor permitted by equation (5.1), with . If the final total weight is positive, its total absolute loss satisfies
The result holds for every permitted factor choice and every finite instance and label sequence. It is the continuous-prediction counterpart of the paper's binary Weighted Majority bound.
Formalization Note The instances themselves are implicit in the supplied expert predictions, since the inequality uses them only through those predictions. The paper's bound is infinite when ; this real-valued statement covers its finite case.
import Definitions.Def_WeightedMajority_Continuous_WMCRun
namespace WeightedMajority.Continuous
/-- Theorem 5.2, p. 236: the total absolute loss of any WMC run. -/
theorem theorem_5_2 {n t : ℕ} (beta : ℝ) (w : ℕ → Fin n → ℝ)
(x : Fin t → Fin n → ℝ) (rho : Fin t → ℝ)
(h : IsWMCRun beta w x rho) (hfin : 0 < WeightedMajority.Basic.totalWeight w t) :
masterLoss w x rho ≤
Real.log (WeightedMajority.Basic.totalWeight w 0 / WeightedMajority.Basic.totalWeight w t) / (1 - beta) := by sorry
end WeightedMajority.Continuous
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.