A Note on Reductions Between Compressed Sensing Guarantees

Abstract

In compressed sensing, one wishes to acquire an approximately sparse high-dimensional signal x∈ℝn via m≪n noisy linear measurements, then later approximately recover x given only those measurement outcomes. Various guarantees have been studied in terms of the notion of approximation in recovery, and some isolated folklore results are known stating that some forms of recovery are stronger than others, via black-box reductions. In this note we provide a general theorem concerning the hierarchy of strengths of various recovery guarantees. As a corollary of this theorem, by reducing from well-known results in the compressed sensing literature, we obtain an efficient ℓp/ℓp scheme for any 0<p<1 with the fewest number of measurements currently known amongst efficient schemes, improving recent bounds of [SomaY16].