GAGA: GPU Accelerated Greedy Algorithms. Welcome to GAGA, a software package for solving large compressed sensing problems with millions of unknowns in fractions of a second by exploiting the power of graphics processing units. The current release GAGA 1.1.0 consists of ten greedy algorithms using five matrix ensembles. This release is set to compile as Matlab executables to enhance your compressed sensing research and applications. A user guide is available for download detailing the capabilities including simple implementations for large-scale testing at problem sizes previously too computationally expensive for extensive testing. The current version, GAGA 1.1.0, contains ten algorithms and is equipped with three clases of matrix multiplication, generic dense matrices, sparse matrices, and the subsampled discrete cosine transform. For large-scale testing, there are a total of five randomly generated matrix ensembles and three randomly generated sparse vector ensembles. For applications, the algorithms are equipped to employ any dense matrix and any sparse matrix in COO format (the default in Matlab). GAGA provides massive acceleration with up to 70x speed-ups in the algorithms’ subroutines over a CPU based matlab implementation. For large scale testing, the GPU based random problem generation can offer up to 1600x acceleration.
Keywords for this software
References in zbMATH (referenced in 2 articles )
Showing results 1 to 2 of 2.
- Blanchard, Jeffrey D.; Tanner, Jared: Performance comparisons of greedy algorithms in compressed sensing. (2015)
- Blanchard, Jeffrey D.; Tanner, Jared: GPU accelerated greedy algorithms for compressed sensing (2013)