Research map / Graph learning / Graph self-supervised learning
Improving Hyperbolic Representations via Gromov-Wasserstein Regularization
Yifei Yang, Wonjun Lee, Dongmian Zou, Gilad Lerman
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Gromov-Monge regularization to hyperbolic ProtoNet | hyperbolic ProtoNet (ResNet18) · MiniImageNet 5-Shot 5-Way | accuracy (%): 68.43 -> 71.64 | Improved |
| add GW regularization to hyperbolic ProtoNet (Conv4) | hyperbolic ProtoNet · CUB 1-Shot 5-Way | accuracy (%): 64.66 -> 68.67 | Improved |
| add GW regularization to hyperbolic ProtoNet (Conv4) | hyperbolic ProtoNet · CUB 5-Shot 5-Way | accuracy (%): 80.29 -> 80.54 | No clear effect |
| add GW regularization to HNN for node classification | HNN · Disease node classification | F1 (%): 58.40 -> 65.62 | Improved |
| add GW regularization to HyboNet for node classification | HyboNet · Wisconsin node classification | F1 (%): 74.07 -> 80.87 | Improved |
| add GW regularization to HGCN for link prediction | HGCN · Cora link prediction | AUC (%): 93.37 -> 93.48 | No clear effect |
| add GW regularization to HGCN for link prediction | HGCN · Airport link prediction | AUC (%): 93.42 -> 95.81 | Improved |
| add GW regularization (training cost) | hyperbolic ProtoNet (ResNet18, MiniImageNet 5-Shot) · per-epoch runtime (s) | runtime: 43.2 -> 49.2 | Hurt |
| add GW regularization to reduce GW distance between HNN input and output features | hyperbolic ProtoNet Conv4 · CUB 5-Shot 5-Way | GW distance: 47833.570 -> 1.575 | Improved |
| vary beta hyperparameter around chosen value (ablation) | GW regularization · Disease-LP and Disease-NC | performance | No clear effect |
Closest papers by meaning
- Random Laplacian Features for Learning with Hyperbolic Space (ICLR 2023 · similarity 0.90)
- Hyperbolic Representation Learning: Revisiting and Advancing (ICML 2023 · similarity 0.89)
- Fully Hyperbolic Convolutional Neural Networks for Computer Vision (ICLR 2024 · similarity 0.88)
- Learning Structured Representations with Hyperbolic Embeddings (NeurIPS 2024 · similarity 0.88)
- Residual Hyperbolic Graph Convolution Networks (AAAI 2024 · similarity 0.88)
- Hyperbolic Embeddings of Supervised Models (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
