Fig. 4.

Comparison between KLIC, COCA and other clustering algorithms. The labels ‘RBF opt.’ and ‘RBF fixed’ refer to the MKL method using an RBF kernel with either σ optimized or fixed at 1 (see text). Top: ARI obtained with each clustering algorithm using four datasets having the same clustering structure and cluster separability (as in Fig. 2). Bottom: ARI obtained with COCA and KLIC for each of the subsets of heterogeneous datasets considered in Figure 3. The high ARI obtained with KLIC in all settings shows the advantage of using this method, especially when some of the datasets are noisy