TR2016-039

Learning optimal nonlinearities for iterative thresholding algorithms


    •  Kamilov, U. S.; Mansour, H., "Learning Optimal Nonlinearities for Iterative Thresholding Algorithms", IEEE Signal Processing Letters, DOI: 10.1109/LSP.2016.2548245, ISSN: 1070-9908, Vol. 23, No. 5, pp. 747-751, March 2016.
      BibTeX Download PDF
      • @article{Kamilov2016mar2,
      • author = {Kamilov, U. S. and Mansour, H.},
      • title = {Learning Optimal Nonlinearities for Iterative Thresholding Algorithms},
      • journal = {IEEE Signal Processing Letters},
      • year = 2016,
      • volume = 23,
      • number = 5,
      • pages = {747--751},
      • month = mar,
      • doi = {10.1109/LSP.2016.2548245},
      • issn = {1070-9908},
      • url = {http://www.merl.com/publications/TR2016-039}
      • }
  • MERL Contact:
  • Research Areas:

    Computational Sensing, Multimedia


Iterative shrinkage/thresholding algorithm (ISTA) is a well-studied method for finding sparse solutions to illposed inverse problems. In this letter, we present a data-driven scheme for learning optimal thresholding functions for ISTA. The proposed scheme is obtained by relating iterations of ISTA to layers of a simple feedforward neural network and developing a corresponding error backpropagation algorithm for fine-tuning the thresholding functions. Simulations on sparse statistical signals illustrate potential gains in estimation quality due to the proposed data adaptive ISTA.