Multi Objective Hyperparameter Optimization

Multi-objective hyperparameter optimization (MO-HPO) addresses the challenge of tuning machine learning models to simultaneously optimize multiple, often conflicting, objectives beyond just predictive accuracy (e.g., speed, fairness, energy efficiency). Current research focuses on improving the efficiency and effectiveness of MO-HPO algorithms, including Bayesian optimization and population-based training methods, often incorporating techniques like surrogate models and meta-learning to accelerate the search process. This field is crucial for developing robust and responsible machine learning systems, enabling practitioners to find optimal model configurations that balance competing performance criteria and address broader societal concerns.