Kim, Seung-JunAbid, Dan2021-01-292021-01-292017-01-0111785http://hdl.handle.net/11603/20747Dictionary learning based on sparse coding has exhibited excellent performance for various tasks such as denoising, prediction and classication with diverse applictions. However, sparse coding-based dictionary learning does not capture potential clusters of subspaces in the data. In this work, dictionary learning based on both sparsity and low rank properties is formulated and ecient solution methods are derived in both batch and online implementations. The algorithms are applied to a spectrum sensing problem for cognitive radios. The numerical experiments illustrate the merit of the novel approach. Furthermore, the algorithm is extended to the spectrum prediction problem, where the future interference levels are forecasted. Finally, a RF signal classciation problem is tackled using a deep layered architecture combining the scattering transform and the convolutional neural network.application:pdfText