作者: Zongming Ma , Andreas Buja , Dan Yang
DOI:
关键词: Iterative thresholding 、 Rank (linear algebra) 、 Mathematics 、 Noise reduction 、 Minimax 、 Large class 、 Algorithm
摘要: We study minimax rates for denoising simultaneously sparse and low rank matrices in high dimensions. show that an iterative thresholding algorithm achieves (near) optimal adaptively under mild conditions a large class of loss functions. Numerical experiments on synthetic datasets also demonstrate the competitive performance proposed method.