eduzhai > Applied Sciences > Engineering >

VPNet Variable Projection Networks

  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: In this paper, we introduce VPNet, a novel model-driven neural networkarchitecture based on variable projections (VP). The application of VPoperators in neural networks implies learnable features, interpretableparameters, and compact network structures. This paper discusses the motivationand mathematical background of VPNet as well as experiments. The concept wasevaluated in the context of signal processing. We performed classificationtasks on a synthetic dataset, and real electrocardiogram (ECG) signals.Compared to fully-connected and 1D convolutional networks, VPNet features fastlearning ability and good accuracy at a low computational cost in both of thetraining and inference. Based on the promising results and mentionedadvantages, we expect broader impact in signal processing, includingclassification, regression, and even clustering problems.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×