Convex and Semi-Nonnegative Matrix Factorizations
read more
Citations
Nonnegative Matrix and Tensor Factorizations: Applications to Exploratory Multi-way Data Analysis and Blind Source Separation
Data-Driven Intelligent Transportation Systems: A Survey
Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation
Algorithms for nonnegative matrix factorization with the β-divergence
Nonnegative Matrix Factorization: A Comprehensive Review
References
Latent dirichlet allocation
Latent Dirichlet Allocation
Learning the parts of objects by non-negative matrix factorization
Learning parts of objects by non-negative matrix factorization
Algorithms for Non-negative Matrix Factorization
Related Papers (5)
Frequently Asked Questions (10)
Q2. What is the objective function for fixing G?
The objective function that the authors minimize is the following sum of squared residuals:J = ||X − FGT ||2 = Tr (XT X − 2XT FGT + GF T FGT ). (13)Fixing G, the solution for F is obtained by computing dJ/dF = −2XG + 2FGTG = 0.
Q3. What are the log messages in the DRAFT?
The log messages are grouped into 9 categories: configuration, connection, create, dependency, other, report, request, start, and stop.
Q4. What is the auxiliary function of the iterative update algorithm?
A function Z(H, H̃) is called an auxiliary functionof J(H) if it satisfiesZ(H, H̃) ≥ J(H), Z(H, H) = J(H), (19)November 5, 2008 DRAFTfor any H, H̃ .
Q5. How can the authors consider shifting mixed-sign data to be nonnegative?
November 5, 2008 DRAFTWhile their algorithms apply directly to mixed-sign data, it is also possible to consider shifting mixed-sign data to be nonnegative by adding the smallest constant so all entries are nonnegative.
Q6. What is the objective function for clustering?
The authors can absorb D − 1 2 n into G and solve forCluster-NMF : X ≈ XG+GT+. (38)The authors call this factorization Cluster-NMF because the degree of freedom in this factorization is the cluster indicator G, as in a standard clustering problem.
Q7. How do the authors measure the sparsity of G in the experiments?
To measure the sparsity of G in the experiments, the authors compute the average of each column of G and set all elements below 0.001 times the average to zero.
Q8. Why do the authors believe that NMF is better than K-means?
The authors believe this is due to the flexibility of matrix factorization as compared to the rigid spherical clusters that the K-means clustering objective function attempts to capture.
Q9. What is the correct solution for the update rule for G?
Theorem 1: (A) Fixing F , the residual ||X−FGT ||2 decreases monotonically (i.e., it is nonincreasing) under the update rule for G. (B) Fixing G, the update rule for F gives the optimal solution to minF ||X − FG||2. Proof.
Q10. What is the description of the NMF algorithm?
The authors also showed that the NMF variants can be viewed as relaxations of K-means clustering, thus providing a closer tie between NMF and clustering than has been present in the literature to date.