Preference Based Optimization

Preference-based optimization tackles the challenge of finding optimal solutions when direct evaluation of an objective function is impossible or impractical, relying instead on pairwise comparisons of candidate solutions. Current research focuses on developing efficient algorithms, such as surrogate-based methods (e.g., using piecewise affine or radial basis function surrogates) and meta-optimization frameworks, to effectively explore the solution space and converge to optimal or near-optimal solutions with minimal comparisons. These techniques find applications in diverse fields, from personalized energy consumption estimation to complex engineering design problems where human preferences or expensive simulations drive the optimization process. The development of robust and theoretically grounded algorithms with convergence guarantees is a key area of ongoing investigation.