Green Optimization: Energy-aware Design of Metaheuristics by Using Machine Learning Surrogates to Cope with Real Problems

Researchers propose using machine learning models as efficient stand-ins to guide complex problem-solving algorithms. This 'green optimization' approach dramatically cuts the computational resources needed. Experiments show a pre-trained AI surrogate can reduce energy use and execution time by up to 98%, and memory usage by 99%. However, the method isn't a universal fix, as it can sometimes hurt accuracy, requiring careful integration with traditional performance metrics.