Bayesian optimization

An efficient, sequential design strategy for optimizing expensive, black-box functions using probabilistic modeling.

Bayesian optimization is the go-to strategy for tuning hyperparameters, designing materials, or configuring complex systems when evaluating a single setup is slow or expensive (think hours of training time or costly lab trials). The framework operates by building a probabilistic surrogate model (typically a Gaussian process) of the target function based on previous evaluations. It then uses an acquisition function (such as Expected Improvement or Upper Confidence Bound) to mathematically balance exploration of uncertain regions with exploitation of known high-performing areas. This loop allows teams to find global optima in dozens of iterations rather than the thousands required by brute-force grid search.

https://botorch.org/docs/overview