Bottom-Up Multiple Row Addition Algorithms for the Biclustering-Problem

The discovery of locally and significantly correlated subpatterns within a two-dimensional dataset has recently become quite popular and is amongst others addressed by methods solving the biclustering problem. The preservation of a particularly defined degree of homogeneity between elements within a bicluster plays a key role in the search procedure. A prominent quantity is known as the mean squared residue. Most approaches use such measures only to evaluate the found solutions a posteriori, instead of incorporating them directly into the search procedure. This work proposes a pairwise distance function related to the mean squared residue and includes this measure into two new multiple enrichment algorithms. The impact is demonstrated empirically by the enrichment of bicluster sets of a popular method and by the enrichment of randomly chosen single rows.