Sparsity Assumption

An assumption in statistical learning that the model parameter vector has only a few nonzero elements, exploitable through ℓ1 regularization or greedy algorithms to reduce the dependence of estimation error on dimensionality.