English

An Approach to One-Bit Compressed Sensing Based on Probably Approximately Correct Learning Theory

Machine Learning 2017-10-24 v1

Abstract

In this paper, the problem of one-bit compressed sensing (OBCS) is formulated as a problem in probably approximately correct (PAC) learning. It is shown that the Vapnik-Chervonenkis (VC-) dimension of the set of half-spaces in Rn\mathbb{R}^n generated by kk-sparse vectors is bounded below by klg(n/k)k \lg (n/k) and above by 2klg(n/k)2k \lg (n/k), plus some round-off terms. By coupling this estimate with well-established results in PAC learning theory, we show that a consistent algorithm can recover a kk-sparse vector with O(klg(n/k))O(k \lg (n/k)) measurements, given only the signs of the measurement vector. This result holds for \textit{all} probability measures on Rn\mathbb{R}^n. It is further shown that random sign-flipping errors result only in an increase in the constant in the O(klg(n/k))O(k \lg (n/k)) estimate. Because constructing a consistent algorithm is not straight-forward, we present a heuristic based on the 1\ell_1-norm support vector machine, and illustrate that its computational performance is superior to a currently popular method.

Keywords

Cite

@article{arxiv.1710.07973,
  title  = {An Approach to One-Bit Compressed Sensing Based on Probably Approximately Correct Learning Theory},
  author = {Mehmet Eren Ahsen and Mathukumalli Vidyasagar},
  journal= {arXiv preprint arXiv:1710.07973},
  year   = {2017}
}

Comments

28 pages, 4 figures

R2 v1 2026-06-22T22:21:54.983Z