The Variational Nystrom method for large-scale spectral problems

Authors: Max Vladymyrov, Miguel Carreira-Perpinan

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4. Experiments To set up the spectral problem (P) we want to approximate, we use Laplacian Eigenmaps (LE) (Belkin & Niyogi, 2003), a spectral manifold learning algorithm. We also use spectral clustering (SC) in an image segmentation experiment. We compare the approximations to the exact embedding X or the ground truth, when available. Fig. 1 (top) shows the relative error with respect to the ground truth as we change the Gaussian affinity bandwidth σ.
Researcher Affiliation Collaboration Max Vladymyrov MXV@GOOGLE.COM Google, Inc. Miguel A. Carreira-Perpi n an MCARREIRA-PERPINAN@UCMERCED.EDU Electrical Engineering and Computer Science, School of Engineering, University of California, Merced
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide an unambiguous statement of code release or a direct link to a source-code repository for the described methodology.
Open Datasets Yes We used 5 000 points from the Swiss roll dataset, for which the ground truth is available. ... We use 20 000 random digits from MNIST and reduce dimensionality to d = 10 using exact LE and each of the methods. ... We used 1 020 000 points from the infinite MNIST dataset (Loosli et al., 2007)
Dataset Splits No The paper does not explicitly provide specific percentages, sample counts, or citations to predefined train/validation/test splits for the datasets, nor does it detail a specific splitting methodology for data partitioning.
Hardware Specification No No specific hardware details (such as GPU/CPU models, processor types, or memory amounts) used for running the experiments were provided in the paper.
Software Dependencies No The paper mentions 'Matlab s eigs routine' but does not specify the version number for Matlab or any other software dependencies with their specific versions.
Experiment Setup Yes To compute X, we use Matlab s eigs routine with default parameters (maxit = 300, tol = eps, p = 2d). ... We construct the affinity matrix using entropic affinities ... with perplexity K = 30 ... We also sparsify the affinity matrix by zeroing all but the 200 largest values in each row.