Projection Robust Wasserstein Barycenters
Authors: Minhui Huang, Shiqian Ma, Lifeng Lai
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We incorporate the RPRWB into a discrete distribution clustering algorithm, and the numerical results on real text datasets confirm that our RPRWB model helps improve the clustering performance significantly. |
| Researcher Affiliation | Academia | 1 Department of Electrical and Computer Engineering, University of California, Davis, CA, USA 2 Department of Mathematics, University of California, Davis, CA, USA. |
| Pseudocode | Yes | Algorithm 1 The RGA-IBP Algorithm; Algorithm 2 The RBCD Algorithm; Algorithm 3 Round(π, p, q) |
| Open Source Code | Yes | Code available at https://github.com/mhhuang95/PRWB. |
| Open Datasets | Yes | We use the pre-trained word-vector dataset Glo Ve (Pennington et al., 2014)... The Reuters Subset is a 5-class subset of the Reuters dataset 2. The BBCnews Abstract and BBCsport Abstract 3 (Greene & Cunningham, 2006)... |
| Dataset Splits | No | The paper describes using datasets for clustering performance evaluation and mentions AMI scores, but does not explicitly detail a validation dataset split or usage. |
| Hardware Specification | Yes | All experiments are conducted on a Linux server with a 32-core Intel Xeon CPU (E5-2667, v4, 3.20GHz per core). |
| Software Dependencies | No | The paper mentions using Glo Ve 300d word vectors but does not specify software dependencies with version numbers for the implementation of the proposed algorithms or other experimental tools. |
| Experiment Setup | Yes | We set the step size τ = 0.0005 for both RBCD and RGA-IBP algorithms and η = 0.5 mid({Cl}l [m])... We set ϵ = 10 4... We choose k = 2 for the BBCsport Abstract dataset and k = 3 for the Reuters Subset and the BBCnews Abstract datasets. |