Related Experiment Video
Updated: Feb 21, 2026
04:57
Assisted Selection of Biomarkers by Linear Discriminant Analysis Effect Size LEfSe in Microbiome Data
Published on: May 16, 2022
17.6K
Detection of Sources in Non-Negative Blind Source Separation by Minimum Description Length Criterion
IEEE Transactions on Neural Networks and Learning Systems
|October 6, 2017
Summary
Determining the number of sources in non-negative blind source separation (nBSS) is challenging. This study introduces a revised Minimum Description Length (MDL) criterion for accurate source number detection in nBSS.
Area of Science:
- Signal Processing
- Machine Learning
- Data Analysis
Background:
- Non-negative blind source separation (nBSS) is widely used but struggles with accurate model order selection (determining the number of sources).
- Existing methods often fail due to oversimplified models, invalid assumptions, and subjective thresholds.
- Previous attempts to improve nBSS model order selection have shown limited success.
Purpose of the Study:
- To develop a formally revised model order selection criterion for nBSS based on the Minimum Description Length (MDL) principle.
- To address the limitations of existing methods by incorporating more realistic assumptions and models.
- To provide a robust and accurate method for determining the number of sources in nBSS problems.
Main Methods:
- Formulated a new model order selection criterion rooted in the Minimum Description Length (MDL) principle.
- Assumed a unique nBSS solution with a deterministic mixing matrix and multivariate Dirichlet distributed source signals.
- Developed a computationally efficient stochastic algorithm for parameter estimation and used Monte Carlo integration for description length calculation.
- Exploited the geometric properties of the data simplex in nBSS for modeling and estimation.
Main Results:
- The proposed nBSS-MDL criterion consistently and accurately detects the true number of sources across extensive simulations.
- Validation on four real-world datasets demonstrates the criterion's strong performance and general applicability.
- The method overcomes limitations of previous approaches, avoiding order over- and under-estimation.
Conclusions:
- The revised nBSS-MDL criterion offers a significant advancement in solving the critical problem of model order selection in nBSS.
- This new criterion provides a reliable and accurate tool for identifying the number of sources in diverse nBSS applications.
- The study highlights the effectiveness of the MDL principle when applied with realistic assumptions for complex signal processing tasks.
