A Nonconvex Projection Method for Robust PCA

Authors: Aritra Dutta, Filip Hanzely, Peter Richtàrik1468-1476

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate through extensive numerical experiments on a variety of applications, including shadow removal, background estimation, face detection, and galaxy evolution, that our approach matches and often significantly outperforms current state-of-the-art in various ways.
Researcher Affiliation Academia Aritra Dutta is with the Visual Computing Center, Division of Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) at King Abdullah University of Science and Technology, Thuwal, Saudi Arabia-23955-6900, e-mail: aritra.dutta@kaust.edu.sa. Filip Hanzely is with the Visual Computing Center, Division of Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) at King Abdullah University of Science and Technology, Thuwal, Saudi Arabia-23955-6900, e-mail: filip.hanzely@kaust.edu.sa. Peter Richt arik is with the Visual Computing Center, Division of Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) at King Abdullah University of Science and Technology, University of Edinburgh, and MIPT, e-mail: peter.richtarik@kaust.edu.sa.
Pseudocode Yes Algorithm 1: Alternating projection method for set feasibility
Open Source Code No The paper does not provide a link or explicit statement for the open-sourcing of its own code.
Open Datasets Yes We used the Basic sequence of the Stuttgart artificial dataset (Brutzer, H oferlin, and Heidemann 2012).
Dataset Splits No The paper describes generating synthetic data and using real-world datasets but does not explicitly provide details about train/validation/test splits or cross-validation for its experiments.
Hardware Specification No The paper does not provide specific details on the hardware used for running experiments.
Software Dependencies No The paper mentions using a 'fast implementation of n-th element computation from (Li 2013)' and sets parameters for algorithms (APG, i EALM, Go Dec) but does not specify any software names with version numbers.
Experiment Setup Yes For both APG and i EALM, we set λ = 1/ m and for i EALM we use µ = 1.25/ A 2 and ρ = 1.5, where A 2 is the spectral norm (maximum singular value) of A. For Go Dec we set q = 2.