The sample complexity of weighted sparse approximation

For Gaussian sampling matrices, we provide bounds on the minimal number of measurements $m$ required to achieve robust weighted sparse recovery guarantees in terms of how well a given prior model for the sparsity support aligns with the true underlying support. Our main contribution is that for a sparse vector ${\bf x} \in \mathbb{R}^N$ supported on an unknown set $\mathcal{S} \subset \{1, \dots, N\}$ with $|\mathcal{S}|\leq k$, if $\mathcal{S}$ has \emph{weighted cardinality} $\omega(\mathcal{S}) := \sum_{j \in \mathcal{S}} \omega_j^2$, and if the weights on $\mathcal{S}^c$ exhibit mild growth, $\omega_j^2 \geq \gamma \log(j/\omega(\mathcal{S}))$ for $j\in\mathcal{S}^c$ and $\gamma > 0$, then the sample complexity for sparse recovery via weighted $\ell_1$-minimization using weights $\omega_j$ is linear in the weighted sparsity level, and $m = \mathcal{O}(\omega(\mathcal{S})/\gamma)$. This main result is a generalization of special cases including a) the standard sparse recovery setting where all weights $\omega_j \equiv 1$, and $m = \mathcal{O}\left(k\log\left(N/k\right)\right)$; b) the setting where the support is known a priori, and $m = \mathcal{O}(k)$; and c) the setting of sparse recovery with prior information, and $m$ depends on how well the weights are aligned with the support set $\mathcal{S}$. We further extend the results in case c) to the setting of additive noise. Our results are {\em nonuniform} that is they apply for a fixed support, unknown a priori, and the weights on $\mathcal{S}$ do not all have to be smaller than the weights on $\mathcal{S}^c$ for our recovery results to hold.