In recent neuroscience research, it is of significant importance to develop an efficient method to predict subject's behaviors or cognitive states from his brain activities for both its applications and methodology. In ATR Computational Neuroscience Laboratories, we have developed "sparse estimation algorithms" and successfully applied them to various problems. There are several merits in sparse estimation algorithms

1. they are applicable for problems with small number of samples and high (more than several thousand) dimensional data,
2. they avoid overfitting to some extent,
3. they make a result more interpretable.

In addition, the sparse estimation algorithms here requires no parameter tuning, thus you can apply these algorithms to wide range of data sets immediately.

In the following link, we collect MATLAB toolboxes for sparse estimation algorithms that have been developed by our group and collaborators. We provide three sparse estimation toolboxes for regression problems and one for classification problems. Please apply one of three regression toolboxes when your behavior variable takes continuous value and the classification toolbox when your behavior variable is a categorical variable. Please try them for your data anyway.