Informed Weighted Non-Negative Matrix Factorization Using αβ-Divergence Applied to Source Apportionment.

Area of Science:

• Data analysis and signal processing
• Matrix factorization techniques
• Environmental science and atmospheric chemistry

Background:

• Non-negative matrix factorization (NMF) is a widely used technique for source separation.
• Existing NMF methods often lack the ability to incorporate prior knowledge effectively.
• Incorporating constraints and prior information can significantly enhance NMF performance.

Purpose of the Study:

• To propose novel informed weighted non-negative matrix factorization (NMF) methods.
• To extend existing NMF approaches by incorporating prior knowledge using an α β -divergence cost function.
• To develop new update rules for NMF that leverage available information for improved parameterization.

Main Methods:

• Development of informed weighted non-negative matrix factorization (NMF) algorithms.
• Utilizing an α β -divergence cost function for NMF.
• Incorporating exact knowledge/boundedness of components and row sum-to-one property into NMF parameterization.
• Derivation of new update rules for the proposed NMF methods.

Main Results:

• The proposed informed weighted NMF methods demonstrate relevance across various operating conditions.
• Experiments on simulated particulate matter mixtures show significant improvements.
• Validation with expert knowledge on a real-world dataset confirms the effectiveness of the approaches.

Conclusions:

• Informed weighted NMF with α β -divergence offers a powerful framework for source separation.
• The developed methods effectively integrate prior information for enhanced NMF performance.
• The approach shows promise for applications in environmental monitoring and source apportionment.