A technique that encourages weight matrices in neural networks to be orthogonal, aiding in reducing redundancy and improving generalization.
Last updated 9 months ago