Non-negative multiple matrix factorization

摘要: Non-negative Matrix Factorization (NMF) is a traditional unsupervised machine learning technique for decomposing matrix into set of bases and coefficients under the non-negative constraint. NMF with sparse constraints also known extracting reasonable components from noisy data. However, tends to give undesired results in case highly data, because information included data insufficient decompose. Our key idea that we can ease this problem if complementary are available could integrate estimation coefficients. In paper, propose novel factorization method called Multiple (NMMF), which utilizes as auxiliary matrices share row or column indices target matrix. The sparseness improved by simultaneously, since provide about We formulate NMMF generalization NMF, then present parameter procedure derived multiplicative update rule. examined both synthetic real experiments. effect appeared performance. confirmed obtained were intuitive thanks