Sparsity constrained model for the semantic features selection

Predicting brain activity associated with concrete concepts has been attracted wide attention in brain imaging studies. The main task is to construct a computational model for the link between the stimuli and the brain image. However, the ordinary regression model cannot make the desired selection among the semantic features due to the small sample size problem. In this paper, we propose a sparsity constrained model to automatically choose the relevant semantic features. Specifically, we explicitly constrain the number of semantic features associated with the individual voxels. The motivation is based on the fact that the responses of a voxel to the stimuli can only be explained by a limited number of neuron activity bases. The experimental results on predicting brain images show the effectiveness of the proposed approach, as well as meaningful representation of the concepts.