Fast Algorithms for Robust PCA via Gradient Descent

Authors: Xinyang Yi, Dohyung Park, Yudong Chen, Constantine Caramanis

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Numerical Results In this section, we provide numerical results and compare the proposed algorithms with existing methods, including the inexact augmented lagrange multiplier (IALM) approach [20] for solving the convex relaxation (1) and the alternating projection (Alt Proj) algorithm proposed in [21]. All algorithms are implemented in MATLAB 4, and the codes for existing algorithms are obtained from their authors. SVD computation in all algorithms uses the PROPACK library.5 We ran all simulations on a machine with Intel 32-core Xeon (E5-2699) 2.3GHz with 240GB RAM.
Researcher Affiliation Academia Xinyang Yi Dohyung Park Yudong Chen Constantine Caramanis The University of Texas at Austin Cornell University {yixy,dhpark,constantine}@utexas.edu yudong.chen@cornell.edu
Pseudocode Yes Algorithm 1 Fast RPCA
Open Source Code Yes Our code is available at https://www.yixinyang.org/code/RPCA_GD.zip.
Open Datasets Yes We use two public benchmarks, the Restaurant and Shopping Mall datasets.6
Dataset Splits No The paper describes the generation of synthetic data and the use of public benchmarks but does not specify exact train/validation/test dataset splits or methodologies for data partitioning.
Hardware Specification Yes We ran all simulations on a machine with Intel 32-core Xeon (E5-2699) 2.3GHz with 240GB RAM.
Software Dependencies No The paper mentions that 'All algorithms are implemented in MATLAB' and 'SVD computation in all algorithms uses the PROPACK library', but it does not specify version numbers for either software.
Experiment Setup Yes Suppose we choose γ = 2 and η = c/σ 1 for any c 1/36. The low-rank part M is given by M = AB , where A, B Rd r have entries drawn independently from a zero mean Gaussian distribution with variance 1/d. For a given sparsity parameter α, each entry of S is set to be nonzero with probability α, and the values of the nonzero entries are sampled uniformly from [ 5r/d, 5r/d].