Efficient multiple incremental computation for Kernel Ridge Regression with Bayesian uncertainty modeling

 Abstract—This study presents an energy-economic approach for incremental/decremental learning based on kernel ridge regression, a frequently used regressor on clouds. To avoid reanalyzing the entire dataset when data change every time, the proposed mechanism supports incremental/decremental processing for both single and multiple samples (i.e., batch processing). Moreover, incremental/decremental analyses in empirical and intrinsic space are also introduced to handle with data matrices with a large number of samples or feature dimensions. At the end of this study, we further the proposed mechanism to statistical Kernelized Bayesian Regression, so that incremental/decremental analyses become applicable. Experimental results showed that the performance in accuracy of the proposed method remained as well as the original nonincremental design. Furthermore, training time and power consumption were significantly reduced. These findings thereby demonstrate the effectiveness of the proposed method.