Support Recovery Analysis for Low Precision Compressive Sensing

Low precision data representation in the context of compressive sensing is shown to yield good performance in sparse signal recovery yet potentially relieving processing ability of computer systems. Support recovery performance, however, is of special interest for certain scientific instruments, e.g., radio astronomy, functional Magnetic Resonance Imaging, X-ray CT imaging. Hence, in this work, we will focus on the rigorous analysis of support recovery performance of compressive sensing algorithms in the low precision scheme, i.e., when all input data is quantized. As a baseline, we start with the recovery performance of low precision Iterative Hard Thresholding [1].

Supervised by Nezihe Merve Gürel


  1. [1] Gurel, N. M., Kara, K., Stonajov, A., Smith, T., Alistarh, D., Puschel, M. & Zhang, C. (2018). Compressive Sensing with Low Precision Data Representation: Theory and Applications. arXiv preprint arXiv:1802.04907.
  2. [2] Blumensath, T. & Davies, M. E. (2010). Normalized Iterative Hard Thresholding: Guaranteed Stability and Performance, IEEE Journal of Selected Topics in Signal Processing.
  3. [3] Gopi, S., Netrapalli, P., Jain, P., & Nori, A (2013). One-bit compressed sensing: Provable support and vector recovery, International Conference on Machine Learning.