Ensemble Clustering with Heterogeneous Transfer Learning

This work introduces a novel approach to ensemble clustering by incorporating transfer learning. We address a clustering problem where, in addition to the data being analyzed, we have access to “similar” labeled data. The datasets may have different feature descriptions. Our method revolves around constructing meta-features that capture the structural characteristics of the data and transferring them from the source domain to the target domain. To define meta-features, we use the cluster ensemble method. Through extensive Monte Carlo modeling experiments, we have demonstrated the effectiveness of our proposed method. Notably, compared to other similar approaches, our method exhibits the capability to handle arbitrary feature descriptions in both the source and target domains. Additionally, it offers a reduced computational complexity, making it more efficient in practice.