Learning Narrowband Graph Spectral Kernels for Graph Signal Estimation Çizge Sinyallerinin Dar Bantli Spektral Kernel Öǧrenimi ile Kestirimi

Furkan Kar O., Turhan G., VURAL E.

30th Signal Processing and Communications Applications Conference, SIU 2022, Safranbolu, Turkey, 15 - 18 May 2022 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu55565.2022.9864966
  • City: Safranbolu
  • Country: Turkey
  • Keywords: graph dictionary learning, graph kernels, Graph signal processing, narrow-band kernels
  • TED University Affiliated: Yes


© 2022 IEEE.In this work, we study the problem of estimating graph signals from incomplete observations. We propose a method that learns the spectrum of the graph signal collection at hand by fitting a set of narrowband graph kernels to the observed signal values. The unobserved graph signal values are then estimated using the sparse representations of the signals in the graph dictionary formed by the learnt kernels. Experimental results on graph data sets show that the proposed method compares favorably to baseline graph-based semi-supervised regression solutions.