Library for Batched version of KSVD written in PyTorch that runs on GPU
This library is a batched version of the pytorch KSVD code https://github.com/nel215/ksvd
https://github.com/nel215/ksvd only works with 2d matrix (A, B). This repo efficiently extends it to (Batch, A, B).
KSVD is a popular sparse dictionary learning algorithm used to decompose W as C*B where C (coefficient) is sparse and B (basis) is dense. Original Paper link: https://csaws.cs.technion.ac.il/~ronrubin/Publications/KSVD-OMP-v2.pdf
example_command.sh contains example of how to run the batched KSVD on GPU using pytorch.
Running example_command.sh computes the coefficient & basis of the given weight matrix & sparsity and saves it with --save_name.npy extension.
This library was written as part of work at MSR India.