FFTLasso: Large-Scale LASSO in the Fourier Domain

Authors: Adel Bibi, Hani Itani and Bernard Ghanem

Project Website: https://ivul.kaust.edu.sa/Pages/pub-fft-lasso.aspx

Personal Website: www.adelbibi.com

License: See LICENSE file

If you use any of this work please cite:
Adel Bibi, Hani Itani, Bernard Ghanem
"FFTLasso: Large-Scale LASSO in the Fourier Domain"
Conference on Computer Vision and Pattern Recognition (CVPR 2017)

The code is tested on windows. The required depdencies are:

a) tensorflow-gpu 1.0.1

b) Jupter notebook

c) scipy 0.19.0

d) numpy 1.12.1

The subdirectory "FFTLasso_TensorFlow_VerticalSplits" demonstrates how FFTLasso can be implemented with arbitrary number of vertical splits of the dictionary A to be distrubted over multiple GPUs.

Note: Tensorflow in general is much slower than having a CUDA version. This is a simple demo of how the proposed approach can trivially parallelize data over multiple GPUs since the operations involved are elementwise and FFTs. Another version of this can be simply carried out for horizontal splits of the dictionary that is discussed in the supplementary material.