Fig. 2From: Scalable non-negative matrix tri-factorizationBlock-wise partitioning of data matrix X using a 2×2 configuration. Latent factors U and V are each partitioned into two blocks. The remaining latent factor S is not partitioned into blocksBack to article page