Evolutionary soft co-clustering: formulations, algorithms, and applications

摘要: We consider the co-clustering of time-varying data using evolutionary methods. Existing approaches are based on spectral learning framework, thus lacking a probabilistic interpretation. overcome this limitation by developing model in paper. The proposed assumes that observed generated via two-step process depends historic co-clusters. This allows us to capture temporal smoothness probabilistically principled manner. To perform maximum likelihood parameter estimation, we present an EM-based algorithm. also establish convergence EM An appealing feature is it leads soft assignments naturally. evaluate method both synthetic and real-world sets. Experimental results show our consistently outperforms prior method. fully exploit impact methods, further systematic application study analysis Drosophila gene expression pattern images. encode spatial information at particular developmental time point into matrix mesh-generation pipeline. then co-cluster embryonic domains genes simultaneously for multiple points Results co-clusters reflect underlying biology.