Dimensionality Reduction for Wasserstein Barycenter
Authors: Zachary Izzo, Sandeep Silwal, Samson Zhou
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Lastly, our experimental results validate the speedup provided by dimensionality reduction while maintaining solution quality. Finally, we present experimental evaluation of our proposed methodology. Our experiments in Section 7 demonstrate that on natural datasets, we can reduce the dimension by 1-2 orders of magnitude while increasing the solution cost by only 5%. |
| Researcher Affiliation | Academia | Zachary Izzo Stanford University zle.izzo@gmail.com Sandeep Silwal MIT silwal@mit.edu Samson Zhou Carnegie Mellon University samsonzhou@gmail.com |
| Pseudocode | Yes | Algorithm 1 Using dimensionality reduction with any algorithm A for computing WB |
| Open Source Code | No | The paper states: "We use the code and default settings from [Ye19] to compute the Wasserstein barycenter". While [Ye19] refers to a GitHub repository, it is external code used by the authors, not the authors' own implementation or open-source release of the methodology described in this paper. |
| Open Datasets | Yes | FACES dataset: This dataset is used in the influential ISOMAP paper and consists of 698 images of faces in dimension 4096 [TSL00]. MNIST dataset: We subsample 104 images from the MNIST test dataset (dimension 784). |
| Dataset Splits | No | The paper does not explicitly provide details about training, validation, or test splits. It mentions forming distributions from datasets but not how these distributions were split for training or validation. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using "Wbc-matlab" from [Ye19] but does not specify the version numbers for MATLAB or any other software libraries or dependencies used in the experiments. |
| Experiment Setup | Yes | We project our datasets in dimensions d ranging from d = 2 to d = 30 and compute the Wasserstein barycenter for p = 2. For FACES, we limit the support size of the barycenter to be at most 5 points in R4096... For MNIST we limit the support size of the barycenter to be at most 40. |