Analysis of the Self Projected Matching Pursuit Algorithm
This work addresses memory efficiency in solving large linear systems, which is incremental as it builds on existing greedy strategies.
The paper tackles the problem of solving large linear systems with reduced memory usage by analyzing the Self Projected Matching Pursuit algorithm, a low-memory implementation of Orthogonal Matching Pursuit, showing it is suitable for well-posed problems.
The convergence and numerical analysis of a low memory implementation of the Orthogonal Matching Pursuit greedy strategy, which is termed Self Projected Matching Pursuit, is presented. This approach renders an iterative way of solving the least squares problem with much less storage requirement than direct linear algebra techniques. Hence, it appropriate for solving large linear systems. The analysis highlights its suitability within the class of well posed problems.