Robust bicluster winner

What does it mean by "winner bicluster" and "robust winner bicluster"?

After applying the biclustering algorithm to the input data matrix, the columns and rows are clustered giving category and voxel clusters respectively. Each intersection of row and column cluster is a row-column subspace called bicluster. For each voxel (row) cluster, we can find the category (column) winner bicluster, defined by the column cluster that gives maximum mean value. Likewise, for each column cluster, we can also have the voxel (row) winner bicluster.

An interesting question to ask is "Is there any biclusters that are both the category and voxel winner?" This can be found by intersecting the category winner bicluster with the voxel winner bicluster. The "survival" clusters are called "robust winner biclusters" because they are both the winner in category and voxel cluster.

(click at the figure for better resolution.)

Figure above shows:

• (top-right) the winner category cluster (column) for each voxel category (row)
• (bottom-left) the winner voxel cluster (row) for each category cluster (column)
• (bottom-right) the robust winner derived from the intersection of the winner category cluster and the winner voxel cluster. The survival clusters are the winner in both row and column, and therefore are "robust".