Large scale, parallel low rank matrix completion for seismic data interpolation

This applications is available only in the software release for members of SINBAD consortium.

This software implements the SPGLR algorithm using block parallel matrix-matrix multiplication, which allows it to be used on large scale problems.

Author: Curt Da Silva (curtd@math.ubc.ca)

Date: February, 2015

Contents

Downloading & Dependencies

The synthetic examples code can be found in the SLIM software release under applications/Processing/LargeScaleLRI.

The code has been tested with Matlab R2013a and requires the Parallel Computing Toolbox.

This code uses the following packages, also found in the tools part of the SLIM software release.

Running & Parallelism

All of the examples and results are produced by the scripts found in this software release under /applications/Processing/LargeScaleLRI/examples. Start matlab from /applications/Processing/LargeScaleLRI to add the appropriate paths.

To run the scripts, follow the instructions in the README.md file enclosed with the code

Functions

The SPGLR code consists of the spgLR.m function _tools/solvers/SPGLR_PAR, which performs the parameter-cooling method by solving the Lasso subproblems, and spgLRobj.m, which performs the distributed computation of the objective and its gradient. The accompanying documentation is in the README.md file.

Examples and results

The examples of large scale missing trace interpolation using these methods can be found in applications/Processing/LargeScaleLRI/examples

Results are of missing-receiver interpolation is shown in spgLR_bgdata_view.m

References

[1] A. Aravkin, et al. "Fast methods for denoising matrix completion formulations, with applications to robust seismic data interpolation"

Acknowledgements

Thanks to our sponsors and NSERC for their financial support.