Fig. 7
From: Structured feature selection using coordinate descent optimization
Groups with different noise levels. X-axis represents the groups where each consective 20 groups have the same noise level represented in the corresponding colorbar at the top of the figure. Y-axis represents the number of iterations needed by BCGD algorithm to optimize the weights of the corresponging group. Easy groups (less noisy) are terminated early while difficult groups (more noisy) are terminated late