Recovery of Coherent Data via Low-Rank Dictionary Pursuit

Authors: Guangcan Liu, Ping Li

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on randomly generated matrices and real motion sequences verify our claims.
Researcher Affiliation Academia Guangcan Liu Department of Statistics and Biostatistics Department of Computer Science Rutgers University Piscataway, NJ 08854, USA gcliu@rutgers.edu Ping Li Department of Statistics and Biostatistics Department of Computer Science Rutgers University Piscataway, NJ 08854, USA pingli@rutgers.edu
Pseudocode Yes Algorithm 1 Matrix Recovery input: Observed data matrix X Rm n. adjustable parameter: λ. 1. Solve for ˆL0 by optimizing the RPCA problem (1.2) with λ = 1/ n1. 2. Estimate the rank of ˆL0 by ˆr0 = #{i : σi > 10 3σ1}, where σ1, σ2, , σn2 are the singular values of ˆL0. 3. Form L0 by using the rank-ˆr0 approximation of ˆL0. That is, L0 = arg min L L ˆL0 2 F, s.t. rank(L) ˆr0, which is solved by SVD. 4. Construct a dictionary ˆA from L0 by normalizing the column vectors of L0: [ ˆA]:,i = [ L0]:,i [ L0]:,i 2 , i = 1, , n, where [ ]:,i denotes the ith column of a matrix. 5. Solve for Z by optimizing the LRR problem (1.3) with A = ˆA and λ = 1/ n1. output: ˆAZ .
Open Source Code No The paper does not provide any explicit statement about open-sourcing its code or a link to a code repository for the methodology described.
Open Datasets Yes We now present our experiment with 11 additional sequences attached to the Hopkins155 [21] database.
Dataset Splits No The paper describes generating random matrices and using corrupted motion sequences from Hopkins155, but does not provide specific train/validation/test split percentages, sample counts, or references to predefined splits for reproducibility.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, or cloud resources) used for running the experiments.
Software Dependencies No The paper mentions methods like RPCA and LRR, but does not provide specific software dependencies or version numbers for any libraries or tools used in the implementation.
Experiment Setup Yes In the experiments of this paper, we consistently set ε = 10 6 X F. and We replace each missed entry with a number from Bernoulli 1 and both using λ = 1/ n1.