Bayesian optimization with kernel ensembles for acoustic source localization

A new arXiv preprint proposes using Bayesian optimization to jointly estimate source location and seabed geoacoustic parameters, a task that normally demands many evaluations of an costly normal-mode propagation model. Instead of a single Gaussian process surrogate, the method combines an ensemble of kernels and selects the next sampling point based on disagreement among them. The authors report accurate parameter estimates while keeping the number of expensive model runs low.