Online evaluation of communication models derived via transfer learning

This paper assesses an online method for testing the plausibility of a transferred Gaussian process (GP) communication model in a new environment based on a minimal number of signal measurements. The transferred GP is constructed by combining hyperparameters learned in the original environment with signal measurements collected by an unmanned aircraft (UA) in the new goal environment. The GP´s performance and plausibility is evaluated on a separate validation set in a quick and computationally efficient manner using decision-theoretic hypothesis testing and risk minimization.