Innovation Pursuit: A New Approach to the Subspace Clustering Problem

Authors: Mostafa Rahmani, George Atia

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The numerical simulations demonstrate that i Pursuit can often outperform the state-of-the-art subspace clustering algorithms more so for subspaces with significant intersections along with substantial reductions in computational complexity.
Researcher Affiliation Academia 1University of Central Florida, Orlando, Florida, USA.
Pseudocode Yes Algorithm 1 Innovation pursuit (i Pursuit) for noisy data
Open Source Code No No explicit statement or link providing concrete access to the source code for the described methodology was found.
Open Datasets Yes We apply i Pursuit to the problem of motion segmentation using the Hopkins155 (Tron & Vidal, 2007) dataset, which contains video sequences of 2 or 3 motions.
Dataset Splits No The paper mentions evaluating performance on a dataset but does not specify the training, validation, and test splits (e.g., percentages, sample counts, or predefined splits).
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts, or cloud instance specifications) used for running experiments were mentioned.
Software Dependencies No No specific ancillary software details, such as library names with version numbers (e.g., Python 3.8, PyTorch 1.9), were provided.
Experiment Setup Yes The number of replicates used in spectral clustering for SSC and LRR is equal to 20. Define the clustering error as the ratio of misclassified points to the total number of data points. The right plot of Fig. 3 shows the clustering error versus the dimension of the intersection. The dimension of intersection varies between 1 and 29. Each point in the plot is obtained by averaging over 40 independent runs. ... It is assumed that D follows Data model 1 with M1 = 100, M2 = 500, N = 6 and {ri}6 i=1 = 15. The dimension of the intersection between the subspaces varies from 0 to 14. Thus, the rank of D ranges from 20 to 90. The Noisy data is modeled as De = D + E, with the elements of E sampled independently from a zero mean Gaussian distribution. Fig. 4 shows the performance of the different algorithms versus the dimension of the intersection for τ = E F D F equal to 1/20, 1/10, 1/5 and 1/2.