Matrix completion via modified schatten 2/3-norm

Abstract Low-rank matrix completion is a hot topic in the field of machine learning. It widely used image processing, recommendation systems and subspace clustering. However, traditional method uses nuclear norm to approximate rank function, which leads only suboptimal solution. Inspired by closed-form formulation $$L_{2/3}$$ L 2 / 3 regularization, we propose new truncated schatten 2/3-norm function. Our proposed regularizer takes full account prior information achieves more accurate approximation Based on this regularizer, low-rank model. Meanwhile, fast efficient algorithm are designed solve In addition, rigorous mathematical analysis convergence provided. Finally, superiority our model investigated synthetic data recommender system datasets. All results show that able achieve comparable recovery performance while being faster than state-of-the-art methods.