Numerically Accurate Hyperbolic Embeddings Using Tiling-Based Models
Authors: Tao Yu, Christopher M. De Sa
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our tiling-based model empirically, and show that it can both compress hyperbolic embeddings (down to 2% of a Poincaré embedding on Word Net Nouns) and learn more accurate embeddings on real-world datasets. In Section 7, evaluate our methods on two different tasks: (1) compressing a learned embedding and (2) learning embeddings on multiple real-world datasets. |
| Researcher Affiliation | Academia | Tao Yu Department of Computer Science Cornell University Ithaca, NY, USA tyu@cs.cornell.edu Christopher De Sa Department of Computer Science Cornell University Ithaca, NY, USA cdesa@cs.cornell.edu |
| Pseudocode | Yes | Algorithm 1 Map Lorentz model to L-tiling model. Algorithm 2 RSGD in the L-tiling model. |
| Open Source Code | Yes | We release our compression code in Julia and learning code in Py Torch publicly for reproducibility. https://github.com/ydtydr/Hyperbolic_Tiling_Compression https://github.com/ydtydr/Hyperbolic_Tiling_Learning |
| Open Datasets | Yes | We evaluate our tiling-based model empirically, and show that it can both compress hyperbolic embeddings (down to 2% of a Poincaré embedding on Word Net Nouns) and learn more accurate embeddings on real-world datasets. Datasets Nodes Edges Bio-yeast[29] 1458 1948 Word Net[14] 74374 75834 Nouns 82115 769130 Verbs 13542 35079 Mammals 1181 6541 Gr-QC[23] 4158 13422. |
| Dataset Splits | No | The paper mentions sampling negative examples during training and using standard metrics, but it does not specify explicit training/validation/test splits, percentages, or validation set usage. |
| Hardware Specification | No | The paper states 'All models were trained in float64 for 1000 epochs,' but provides no specific details regarding the hardware used (e.g., GPU models, CPU types, or cloud resources). |
| Software Dependencies | No | The paper mentions using 'Julia' and 'Py Torch' for code release, but it does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | We randomly sample |N(u)| = 50 negative examples per positive example during training. All models were trained in float64 for 1000 epochs with the same hyper-parameters. |