In the last decade, the demand for statistical and computationmethods for data analysis that involve sparse matrices has grown dramatically. The main reason for this is that the classical approaches produce solutions in a form of linear combinations of all variables involved in the problem. However, the nowadays applications deal with huge data sets and the interpretation of linear combinations of tens of thousands of variables is virtually an impossible task. The natural escape is to modify the standard techniques to produce sparse solutions which involve only few of the original variables, but still providing competitive goodness-of-fit to the data. Another reason is the increasing number of problems for analysis of sparse data where a portion of the data entries are missing or grossly corrupted. Such problems require modification of the standard approaches to produce robust solutions which, in turn, may also need to be sparse.
展开▼
