Fast Nonnegative Matrix Factorization Algorithms Using Projected Gradient Approaches for Large-Scale Problems
2008

Fast Nonnegative Matrix Factorization Algorithms Using Projected Gradient Approaches

Sample size: 1000 publication Evidence: moderate

Author Information

Author(s): Rafal Zdunek, Andrzej Cichocki

Primary Institution: Wroclaw University of Technology

Hypothesis

The study investigates the applicability of recent projected gradient methods to nonnegative matrix factorization (NMF) problems.

Conclusion

The proposed NMF projected gradient algorithms are more efficient and consistent than the widely used Lin-PG algorithm.

Supporting Evidence

  • The proposed algorithms were tested on a synthetic benchmark of 4 partially dependent nonnegative signals.
  • The NMF-PSESOP algorithm showed the best performance in terms of signal-to-interference ratio (SIR).
  • Results indicate that the multilayer technique improves performance and consistency across algorithms.

Takeaway

This study looks at ways to break down big data into smaller parts using special math methods, making it faster and easier to understand.

Methodology

The study compares various projected gradient algorithms for NMF using a benchmark of mixed partially dependent nonnegative signals.

Limitations

The algorithms may get stuck in local minima due to the nonconvex nature of the problem.

Digital Object Identifier (DOI)

10.1155/2008/939567

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication