Geometric structure of graph Laplacian embeddings

Authors: Nicolás García Trillos, Franca Hoffmann, Bamdad Hosseini

JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We then generate a random graph by drawing n = 211 vertices from ρ and construct the graph Laplacian matrix n as in (5) by setting ε = 2(log n)3/4 n1/2 0.2. We then construct the discrete Laplacian embedding Fn using the first two eigenvectors of n and compute the number noutside of embedded points that fall outside the cones... We choose |γ| {0.7, 0.8, 0.9, 1, 1.1, 1.2} and repeat the experiment over 20 trials for each value of |γ| where the vertices are redrawn from the mixture. We report the averaged values of noutside/n, as an empirical approximation to the probability mass outside the cones, versus |γ|2 in Figure 2(a) indicating that indeed log(noutside/n) = O(|γ|2)) as expected.
Researcher Affiliation Academia Nicol as Garc ıa Trillos EMAIL Department of Statistics University of Wisconsin-Madison Madison, WI 53706, USA Franca Hoffmann EMAIL Bamdad Hosseini EMAIL Computing and Mathematical Sciences California Institute of Technology Pasadena, CA 91125, USA
Pseudocode No The paper provides mathematical definitions, theorems, and proofs, but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements or links indicating that source code for the methodology is openly available.
Open Datasets No The paper describes generating synthetic data from defined distributions (e.g., 'mixture of two standard Gaussian densities', 'probability measure ρ on a dumbbell shaped domain'). It does not provide concrete access information (links, DOIs, repositories) for pre-existing or specifically generated public datasets.
Dataset Splits No The paper's numerical experiments involve drawing random samples (e.g., 'n = 211 vertices from ρ') and repeating trials, but it does not describe dataset splits for training, validation, or testing.
Hardware Specification No The paper does not specify any hardware details (e.g., GPU/CPU models, memory) used for running its experiments or simulations.
Software Dependencies No The paper does not mention any specific software packages, libraries, or their version numbers used in the experimental setup.
Experiment Setup Yes We then generate a random graph by drawing n = 211 vertices from ρ and construct the graph Laplacian matrix n as in (5) by setting ε = 2(log n)3/4 n1/2 0.2. We then construct the discrete Laplacian embedding Fn using the first two eigenvectors of n and compute the number noutside of embedded points that fall outside the cones C((1, 1)T , π/4 1/4, 0) and C((1, 1)T , π/4 1/4, 0). We choose |γ| {0.7, 0.8, 0.9, 1, 1.1, 1.2} and repeat the experiment over 20 trials for each value of |γ| where the vertices are redrawn from the mixture. ... We let ℓ= 1 and consider dumbbell shaped domains with parameter ϑ {.05, .1, .15, .2, .25, .3, .4, .5}. We then let ρ denote the uniform measure on the dumbbell shaped domain M and generate random graphs with n = 211 vertices sampled randomly from ρ. We choose the rest of the parameters in this experiment identical to the experiment in Example 1 and report the averaged values of noutside/n versus ϑ over 20 trials.