Fig. 4.

Simulation results are for p = 10,000 and $r=\lceil \sqrt{\log (p)}\rceil =4$. Sparsity level k is indicated below each plot. In each plot, the empirical risk of each method [GLRT (triangles); Higher Criticism (diamonds); Max Test (stars)] is plotted against t which corresponds to $A=\sqrt{\frac{\max \{2({\rho }_{\mathrm{logistic}}^{\ast }(\alpha )+t),0\}\log (p)}{r}}$.