Error Bound Conditions and Linear Convergence
摘要
In previous chapters, we leveraged the strong convexity of objective functions to establish convergence rates, particularly linear convergence, for various algorithms. However, strong convexity is often too restrictive in practical applications. To address this limitation, we introduce weaker conditions that still ensure linear convergence. Specifically, we present various error bound conditions, explore their equivalence in the convex setting, and demonstrate how they can be used to establish linear convergence for certain algorithms.