Instead of picking one optimizer upfront, AMM continuously diagnoses problem properties during optimization and routes budget to the best-suited algorithm, outperforming fixed choices and single-algorithm baselines on most black-box continuous optimization tasks.
AMM is an optimizer router that automatically selects and allocates computational budget among five different optimization algorithms based on real-time probes of the problem landscape. It detects features like curvature, plateaus, and fractal structure to match the right optimizer to each problem, achieving the best performance on 64% of tested benchmark problems.