subspace networks

A type of neural network that constrains weight vectors to lie in a low-dimensional subspace, which can help improve generalization performance and reduce overfitting.