Power-Law Entropy Search for Hyperparameter Scaling Laws

"Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search." Meta (Sebastian Ament, David Eriksson, Eytan Bakshy) with NUS proposes a multi-fidelity Bayesian-optimization acquisition that spends compute where it most reduces uncertainty in the fitted scaling law for optimal hyperparameters, rather than in the hyperparameters themselves.