This matlab toolbox propose a generic solver for proximal gradient descent in the convex or non-convex case. It is a complete reimplementation of the GIST algorithm proposed in  with new regularization terms such as the lp pseudo-norm with p=1/2.
When using this toolbox in your research works please cite the paper Non-convex regularization in remote sensing:
D. Tuia, R. Flamary and M. Barlaud, "Non-convex regularization in remote sensing", IEEE transactions Transactions on Geoscience and Remote Sensing, (to appear) 2016.
The code solve optimization problems of the form:
min_x f(x)+lambda g(x)
We provide solvers for solving the following data fitting terms f(x) problems:
The regularization terms g(x) that have been implemented include:
New regularization terms can be easily implemented as discussed in section 3.
All the functions in the toolbox a given in the folder /utils.
The unmix folder contains code and data downloaded from the website of Jose M. Bioucas Dias.
In order to use the function we recommend to execute the following command
if you are not working in the root folder of the toolbox or replacing '.' by the location of the folder on your machine.
We recommend to look at the following files to see how to use the toolbox:
All the regularization terms (and theri proximal operators) are defined in the function utils/get_reg_prox.m.
If you want to add a regularization term (or a projection), you only need to add a case to the switch beginning line 37 and define two functions:
note that in order to limit the number of files, the loss and proximal operators functions are all implemented as subfunctions of file utils/get_reg_prox.m.
You can easily change the data fitting term by providing a new loss and gradient functions to the optimization function utils/gist_opt.m.
A good starting point is by looking at the least square implementation in utils/gist_least.m. Changing the data fitting term correspond to only code the loss function at Line 63 and the corresponding gradient function at Line 59.
We want to thank Jose M. Bioucas Dias for providing the unmixing dataset and functions on his website.
 Gong, P., Zhang, C., Lu, Z., Huang, J., & Ye, J. (2013, June). A General Iterative Shrinkage and Thresholding Algorithm for Non-convex Regularized Optimization Problems. In ICML (2) (pp. 37-45).
 Candes, E. J., Wakin, M. B., & Boyd, S. P. (2008). Enhancing sparsity by reweighted ? 1 minimization. Journal of Fourier analysis and applications, 14(5-6), 877-905.
 Xu, Z., Chang, X., Xu, F., & Zhang, H. (2012). L1/2 regularization: A thresholding representation theory and a fast solver. IEEE Transactions on neural networks and learning systems, 23(7), 1013-1027.