A Probabilistic Graph Coupling View of Dimension Reduction
Authors: Hugues Van Assel, Thibault Espinasse, Julien Chiquet, Franck Picard
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Figure 1: Left: MNIST t-SNE (perp : 30) embeddings initialized with i.i.d N(0, 1) coordinates. Middle: using these t-SNE embeddings, mean coordinates for each digit are represented. Right: we compute a matrix of mean input coordinates for each of the 10 digits and embed it using PCA. For t-SNE embeddings, the positions of clusters vary accross diļ¬erent runs and don t visually match the PCA embeddings of input mean vectors (right plot). |
| Researcher Affiliation | Academia | Hugues Van Assel UMPA ENS Lyon hugues.van_assel@ens-lyon.fr Thibault Espinasse Institut Camille Jordan Lyon 1 Inria Dracula espinasse@math.univ-lyon1.fr Julien Chiquet Agro Paris Tech INRAE julien.chiquet@inrae.fr Franck Picard ENS Lyon franck.picard@ens-lyon.fr |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. |
| Open Source Code | Yes | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] In appendix B. |
| Open Datasets | Yes | MNIST dataset of 10000 samples [14] |
| Dataset Splits | No | The paper uses the MNIST dataset but does not explicitly provide specific train/validation/test splits with percentages or counts in the main text. It mentions 'Other perplexity values for cc PCA are explored in appendix B.2 while the experimental setup is detailed in appendix B.1.', indicating details might be in the appendix, but not directly in the main body for reproducibility. |
| Hardware Specification | No | In appendix B.1, we also include details on the hardware used, along with total runtime. |
| Software Dependencies | No | t-SNE was trained during 1000 iterations using default parameters with the open TSNE implementation [34]. The paper mentions 'open TSNE implementation [34]' but does not provide a specific version number. It also cites other libraries like 'scikit-learn [32]' and 'igraph [13]' without version numbers. |
| Experiment Setup | Yes | t-SNE was trained during 1000 iterations using default parameters with the open TSNE implementation [34]. Other perplexity values for cc PCA are explored in appendix B.2 while the experimental setup is detailed in appendix B.1. |