Over-Relaxation Policies

How it works

Comparisons

Highlighted claims

• The learned policies update the relaxation matrix every ten ADMM iterations without requiring matrix refactorization. — Learning Over-Relaxation Policies for ADMM with Convergence Guarantees
• The scalar policy predicts one relaxation value from global residual and penalty features. — Learning Over-Relaxation Policies for ADMM with Convergence Guarantees
• The vector policy predicts per-constraint relaxation values using shared row-wise weights and per-constraint features. — Learning Over-Relaxation Policies for ADMM with Convergence Guarantees
• Vector policies can exploit per-constraint information but may lose runtime gains because of per-row inference overhead. — Learning Over-Relaxation Policies for ADMM with Convergence Guarantees
• The paper learns low-dimensional solver hyperparameters rather than high-dimensional perturbations to the optimization iteration map. — Learning Over-Relaxation Policies for ADMM with Convergence Guarantees
• Scalar policies are cheaper to evaluate and often produce the best wall-clock runtime. — Learning Over-Relaxation Policies for ADMM with Convergence Guarantees