Improved Graph Laplacian via Geometric Self-Consistency
Authors: Dominique Joncas, Marina Meila, James McQueen
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experimental Results, Synthethic Data. We experimented with estimating the bandwidth ˆϵ on data sampled from two known manifolds..., Semi-supervised Learning (SSL) with Real Data. In this set of experiments, the task is classification on the benchmark SSL data sets proposed by [28]. |
| Researcher Affiliation | Collaboration | Dominique C. Perrault-Joncas Google, Inc. dominiquep@google.com Marina Meil a Department of Statistics University of Washington mmp2@uw.edu James Mc Queen Amazon jmcq@amazon.com |
| Pseudocode | Yes | Algorithm 1 Riemannian Metric(X, i, L, pow { 1, 1}) Algorithm 2 Tangent Subspace Projection(X, w, d ) Algorithm 3 Compute Distortion(X, ϵ, d ) |
| Open Source Code | No | The paper does not provide any links to open-source code or explicitly state that the code for its methodology is publicly available. |
| Open Datasets | Yes | Semi-supervised Learning (SSL) with Real Data. In this set of experiments, the task is classification on the benchmark SSL data sets proposed by [28]. (Digit1, USPS, COIL, BCI, g241c, g241d) |
| Dataset Splits | Yes | We split the training set (consisting of 100 points in all data sets) into two equal groups;5 we minimize the highly non-smooth CV classification error by simulated annealing. In other words, we do 2-fold CV. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as CPU or GPU models, or memory specifications. |
| Software Dependencies | No | The paper does not provide specific software dependencies, such as library names with version numbers, required to replicate the experiments. |
| Experiment Setup | Yes | The range of ϵ values was delimited by ϵmin and ϵmax. We set ϵmax to the average of ||xi xj||2 over all point pairs and ϵmin to the limit in which the heat kernel W becomes approximately equal to the unit matrix; this is tested by maxj(P i Wij) 1 < γ4 for γ 10 4. This range spans about two orders of magnitude in the data we considered, and was searched by a logarithmic grid with approximately 20 points. We saved computatation time by evaluating all pointwise quantities ( ˆD, local SVD) on a random sample of size N = 200 of each data set. |