Robust Subspace Clustering via Thresholding Ridge Regression

Authors: Xi Peng, Zhang Yi, Huajin Tang

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental studies show that TRR outperforms the state-of-the-art methods with respect to clustering quality, robustness, and time-saving. In this section, we investigate the performance of TRR for robust face clustering with respect to clustering quality, robustness, and computational efficiency.
Researcher Affiliation Academia 1Institute for Infocomm Research, Agency for Science, Technology and Research (A*STAR), Singapore 138632 2College of Computer Science, Sichuan University, Chengdu 610065, P.R. China.
Pseudocode Yes Algorithm 1: Robust Subspace Clustering via Thresholding Ridge Regression
Open Source Code Yes 1The codes can be downloaded at the authors website http://www.machineilab.org/users/pengxi/.
Open Datasets Yes We used two popular facial databases, i.e., Extended Yale Database B (Georghiades, Belhumeur, and Kriegman 2001) (Ex Yale B) and AR database (Martinez and Benavente 1998).
Dataset Splits No The paper mentions selecting 'a half of samples to corrupt' for specific tests and analyzing 'the first L subjects', but it does not provide specific train/validation/test split percentages, sample counts, or citations to predefined splits for general model reproduction.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running experiments.
Software Dependencies No The paper mentions software like Node XL and k-means clustering but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes For the SSC algorithm, we experimentally found an optimal α from 1 to 50 with an interval of 1. For LRR, the optimal λ was found from 10-6 to 10 as suggested in (Liu et al. 2013). For LSR and TRR, the optimal λ was chosen from 10-7 to 1. Moreover, a good k was found from 3 to 14 for TRR and from 1 to 100 for LLE-graph. TRR has two parameters, the balance parameter λ and the thresholding parameter k. The values of these parameters depend on the data distribution.