Section: Software and Platforms

SPArse Modeling Software (SPAMS)

Participants : Julien Mairal [correspondant] , Jean-Paul Chieze [WILLOW Project-Team] , Jean Ponce [WILLOW Project-Team] , Francis Bach [SIERRA Project-Team] .

SPAMS v2.4 was released as open-source software in December 2013 (v1.0 was released in September 2009). It is an optimization toolbox implementing algorithms to address various machine learning and signal processing problems involving

• Dictionary learning and matrix factorization (NMF, sparse PCA, ...);

• Solving medium-scale sparse decomposition problems with LARS, coordinate descent, OMP, SOMP, proximal methods;

• Solving large-scale sparse estimation problems with stochastic optimization;

• Solving structured sparse decomposition problems (sparse group lasso, tree-structured regularization, structured sparsity with overlapping groups,...).

The software and its documentation are available at http://spams-devel.gforge.inria.fr/ .

This year, we added new functionalities to the toolbox. A graphical tool for visualizing dictionaries was developed by Jean-Paul Chieze, and stochastic optimization tools corresponding to the papers [24] , [23] were added for dealing with large-scale sparse estimation problems.