P

Initializing...

Theorem 14.11: for λ-strongly convex f, E‖vₜ‖² ≤ ρ² and w⋆ ∈ H, SGD with ηₜ = 1/(λt) and projections onto H has E[f(w̄)] − f(w⋆) ≤ (ρ²/(2λT))(1 + log T) · Prove2Me