Multiple Kernel Testing for SVM-based System Identification

We apply methods of multiple kernel learning to the problem of system identification for multi-dimensional temporal data. Rather than building a full probabilistic model, we take a computationally simple approach that uses out of the box machine learning methods. We attempt to learn the covariance function of a stochastic process via multiple kernel learning. We achieve promising preliminary results and the work suggests an abundance of future theoretical work. We hope to draw on the theory of SVM methods to give a principled learning theory style description of system identification in stochastic processes.