Empirically Measuring Transfer Distance for System Design and Operation

Classical machine learning approaches are sensitive to non-stationarity. Transfer can address non-stationarity by sharing knowledge from one system another, however, in areas like prognostics and defense, data is fundamentally limited. Therefore, transfer algorithms have little, if any, examples which learn. Herein, we suggest that these constraints on algorithmic be addressed systems engineering. We formally define distance general terms demonstrate its use empirically quantifying the transferability of models. consider design rebuild procedures allow for transferable prognostic also predicting operational performance computer vision. Practitioners presented methodology operate with consideration theoretic challenges faced component systems.