Deep transform and metric learning network: Wedding deep dictionary learning and neural network

On account of its many successes in inference tasks and imaging applications, Dictionary Learning (DL) related sparse optimization problems have garnered a lot research interest. In DL area, most solutions are focused on single-layer dictionaries, whose reliance handcrafted features achieves somewhat limited performance. With the rapid development deep learning, improved methods called Deep (DDL), been recently proposed an end-to-end flexible solution with much higher The DDL techniques have, however, also fallen short number issues, namely, computational cost difficulties gradient updating initialization. While few differential programming to speed-up DL, none them could ensure efficient, scalable, robust for methods. To that end, we propose herein, novel differentiable approach, which yields competitive reliable solution. method jointly learns transforms metrics, where each layer is theoretically reformulated as combination one linear Recurrent Neural Network (RNN). RNN shown flexibly layer-associated approximation together learnable metric. Additionally, our work unveils new insights into (NN) DDL, bridging combinations layers Extensive experiments image classification carried out demonstrate can not only outperform existing several counts including, efficiency, scaling discrimination, but achieve better accuracy increased robustness against adversarial perturbations than CNNs.