Lower Bounds for Adaptive Sparse Recovery (1205.3518v2)

Abstract: We give lower bounds for the problem of stable sparse recovery from /adaptive/ linear measurements. In this problem, one would like to estimate a vector $x \in \R^n$ from $m$ linear measurements $A_1x,..., A_mx$. One may choose each vector $A_i$ based on $A_1x,..., A_{i-1}x$, and must output $x*$ satisfying |x* - x|p \leq (1 + \epsilon) \min{k\text{-sparse} x'} |x - x'|_p with probability at least $1-\delta>2/3$, for some $p \in {1,2}$. For $p=2$, it was recently shown that this is possible with $m = O(\frac{1}{\epsilon}k \log \log (n/k))$, while nonadaptively it requires $\Theta(\frac{1}{\epsilon}k \log (n/k))$. It is also known that even adaptively, it takes $m = \Omega(k/\epsilon)$ for $p = 2$. For $p = 1$, there is a non-adaptive upper bound of $\tilde{O}(\frac{1}{\sqrt{\epsilon}} k\log n)$. We show: * For $p=2$, $m = \Omega(\log \log n)$. This is tight for $k = O(1)$ and constant $\epsilon$, and shows that the $\log \log n$ dependence is correct. * If the measurement vectors are chosen in $R$ "rounds", then $m = \Omega(R \log^{1/R} n)$. For constant $\epsilon$, this matches the previously known upper bound up to an O(1) factor in $R$. * For $p=1$, $m = \Omega(k/(\sqrt{\epsilon} \cdot \log k/\epsilon))$. This shows that adaptivity cannot improve more than logarithmic factors, providing the analog of the $m = \Omega(k/\epsilon)$ bound for $p = 2$.