Research map / Graph learning / Graph algorithms and clustering
Overlapping Spaces for Compact Graph Representations
Kirill Sergeevich Shevkunov, Liudmila Prokhorenkova
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 |
|---|---|---|---|
| increase overlapping tree depth from t=0 to t=1 (ablation) | O_l1 · CS PhDs distortion | distortion: 0.0368 -> 0.0300 | Improved |
| increase depth from t=1 to t=3 (ablation) | O_l1 · USCA312 distortion | distortion: 0.00325 -> 0.00257 | Improved |
| increase depth from t=1 to t=3 (ablation) | O_l2 · distortion, USCA312 vs CS PhDs | distortion: 0.00530 -> 0.00303; 0.0328 -> 0.0343 | Mixed |
| add OS-Mixed (weighted inner product) measure to O_l1 t=1 | O_l1, t=1 · Facebook distortion | distortion: 0.0371 -> 0.0172 | Improved |
| add OS-Mixed measure to O_l2 t=1 | O_l2, t=1 · Facebook mAP | mAP: 0.8326 -> 0.9021 | Improved |
| replace exp(-d) with 1/min(d,d0) in proxy-loss (ablation) | O_l2, t=1 · CS PhD mAP | mAP: 0.994 -> 0.810 | Hurt |
| replace exp(-d) with exp(1/min(d,d0)) in proxy-loss (ablation) | H_2^5 product · USCA312 vs CS PhD mAP | mAP: 0.936 -> 0.896; 0.967 -> 0.998 | Mixed |
| use dot product instead of O_l1 t=0 for 256-dim DSSM | DSSM · Wikipedia search 256-dim test mAP | test mAP: 0.677 -> 0.738 | Improved |
| use O_l1 t=0 instead of dot product for 10-dim DSSM | DSSM · Wikipedia search 10-dim test mAP | test mAP: 0.4194 -> 0.4562 | Improved |
| replace OS-Mixed l2 with WIPS on synthetic bipartite graph (stability) | distortion training · synthetic bipartite graph WIPS distortion std | distortion std: 0.0078 -> 0.0018 | Improved |
| use OS-Mixed l2 instead of dot product on synthetic bipartite graph | c-dot · synthetic bipartite graph | distortion: 0.079 -> 0.070 | Improved |
| use best overlapping metric space instead of best product space | O_l1 t=0 · synthetic bipartite graph mAP | mAP: 0.799 -> 0.824 | Improved |
Closest papers by meaning
- Learning Mixed-Curvature Representations in Product Spaces (ICLR 2019 · similarity 0.90)
- Symmetric Spaces for Graph Embeddings: A Finsler-Riemannian Approach (ICML 2021 · similarity 0.88)
- Tight and fast generalization error bound of graph embedding in metric space (ICML 2023 · similarity 0.87)
- Weighted Embeddings for Low-Dimensional Graph Representation (AAAI 2025 · similarity 0.86)
- From Topology to Retrieval: Decoding Embedding Spaces with Unified Signatures (NeurIPS 2026 · similarity 0.85)
- Aligning Embeddings and Geometric Random Graphs: Informational Results and Computational Approaches for the Procrustes-Wasserstein Problem (NeurIPS 2024 · similarity 0.84)
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.
