Multi-Objective Optimization

The field studies problems where improving one objective—such as accuracy, cost, safety, energy use, or fairness—can worsen another, and represents the alternatives as a Pareto frontier. Common approaches include evolutionary algorithms such as NSGA-II, Bayesian optimization for expensive evaluations, gradient-based methods for learning systems, and reinforcement-learning or generative methods for sequential and design tasks. Researchers also develop preference elicitation and constraint-handling methods so people can select useful trade-offs without always computing an entire frontier. A central tension is balancing broad, reliable coverage of feasible trade-offs against evaluation cost, while scalarization, objective scaling, and changing preferences can all affect which solutions are found.