We study the computational cost of recovering a unit-norm sparse principal
component $x \in \mathbb{R}^n$ planted in a random matrix, in either the Wigner
or Wishart spiked model (observing either $W + \lambda xx^\top$ with $W$ drawn
from the Gaussian orthogonal ensemble, or $N$ independent samples from
$\mathcal{N}(0, I_n + \beta xx^\top)$, respectively). P