- Abstract: In representation learning (RL), how to make the learned representations easy to interpret and less overfitted to training data are two important but challenging issues. To address these problems, we study a new type of regularization approach that encourages the supports of weight vectors in RL models to have small overlap, by simultaneously promoting near-orthogonality among vectors and sparsity of each vector. We apply the proposed regularizer to two models: neural networks (NNs) and sparse coding (SC), and develop an efficient ADMM-based algorithm for regularized SC. Experiments on various datasets demonstrate that weight vectors learned under our regularizer are more interpretable and have better generalization performance.
- TL;DR: We propose a new type of regularization approach that encourages non-overlapness in representation learning, for the sake of improving interpretability and reducing overfitting.
- Keywords: Less-overlapness, regularization, near-orthogonality, sparsity