Black Box Optimization Problem

Black-box optimization tackles the challenge of finding optimal solutions for functions whose internal workings are unknown, focusing on efficiently exploring the solution space with minimal evaluations. Current research emphasizes Bayesian optimization, often employing Gaussian processes or tree-based ensembles as surrogate models, and explores advancements in parallel and multi-objective optimization techniques, including the use of transformers and large language models. These methods are crucial for diverse applications like compiler autotuning, hyperparameter optimization in machine learning, and engineering design, where evaluating the objective function is computationally expensive or impractical. The development of standardized benchmarks and improved algorithm selection strategies are also active areas of investigation to enhance reproducibility and efficiency.