A Weak Selection Stochastic Gradient Matching Pursuit Algorithm.

Area of Science:

• Signal Processing
• Machine Learning

Background:

• Existing stochastic gradient matching pursuit algorithms suffer from suboptimal preliminary atomic sets, leading to reduced reconstruction performance and increased computational load.
• These limitations hinder the effective recovery of signals, particularly in complex data types.

Purpose of the Study:

• To address the limitations of current algorithms by proposing a novel method for signal reconstruction.
• To enhance the accuracy and efficiency of signal recovery for various signal types.

Main Methods:

• A weak selection threshold method is introduced to identify the most relevant atoms for the preliminary set.
• The algorithm incorporates a loop termination condition based on the stability of the candidate atomic set size.
• A final check ensures the candidate atomic set dimensions are compatible with the measurement matrix before signal transition estimation.

Main Results:

• The proposed method demonstrates superior reconstruction performance compared to existing stochastic gradient algorithms.
• Effectiveness was validated across diverse signal types: one-dimensional sparse signals, two-dimensional image signals, and low-rank matrix signals.

Conclusions:

• The novel approach significantly improves signal reconstruction capabilities.
• The method offers a more efficient and accurate solution for recovering sparse, image, and low-rank signals.