Greedy Subspace Clustering

Authors: Dohyung Park, Constantine Caramanis, Sujay Sanghavi

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on synthetic data generated from the standard unions of subspaces model demonstrate our theory. We also show that our algorithm performs competitively against state-of-the-art algorithms on realworld applications such as motion segmentation and face clustering, with much simpler implementation and lower computational cost.
Researcher Affiliation Academia Dohyung Park Dept. of Electrical and Computer Engineering The University of Texas at Austin dhpark@utexas.edu Constantine Caramanis Dept. of Electrical and Computer Engineering The University of Texas at Austin constantine@utexas.edu Sujay Sanghavi Dept. of Electrical and Computer Engineering The University of Texas at Austin sanghavi@mail.utexas.edu
Pseudocode Yes Algorithm 1 Nearest Subspace Neighbor (NSN) Input: A set of N samples Y = {y1, . . . , y N}, The number of required neighbors K, Maximum subspace dimension kmax. Output: A neighborhood matrix W 2 {0, 1}N N
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of the code for the methods (NSN+GSR, NSN+Spectral) described in this paper.
Open Datasets Yes For the motion segmentation, we used Hopkins155 dataset [17], which contains 155 video sequences of 2 or 3 motions. For the face clustering, we used Extended Yale B dataset with cropped images from [5, 13].
Dataset Splits No The paper does not explicitly provide specific training/validation/test dataset splits (e.g., percentages, sample counts, or explicit instructions for generating them) for reproduction.
Hardware Specification No The paper mentions running experiments 'on a single desktop' but does not provide specific details about the hardware components (e.g., CPU, GPU models, memory).
Software Dependencies No The paper mentions that MATLAB codes for comparative algorithms were used, but it does not specify specific software dependencies with version numbers for the authors' proposed algorithms (NSN+GSR, NSN+Spectral).
Experiment Setup Yes For NSN, we used the fast implementation described in Section A.1. The numbers of replicates in K-means, K-flats, and the K-means used in the spectral clustering are all fixed to 10. NSN with K = kmax = d followed by GSR with arbitrarily small .