Research map / Graph learning / Graph algorithms and clustering
Beyond Vector Spaces: Compact Data Representation as Differentiable Weighted Graphs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Represent data as a learned weighted graph instead of a low-dimensional Euclidean or Poincare embedding | PRODIGE · MNIST10k, GLOVE10k, and CelebA10k, ≤ 4 parameters per instance | mean squared error: MNIST10k: 0.04683 -> 0.00938; GLOVE10k: 0.11386 -> 0.03289; CelebA10k: 0.01649 -> 0.00374 | Helped |
| Represent data as a learned weighted graph instead of a low-dimensional Euclidean or Poincare embedding | PRODIGE · MNIST10k, GLOVE10k, and CelebA10k, ≤ 8 parameters per instance | mean squared error: MNIST10k: 0.01503 -> 0.00886; GLOVE10k: 0.04839 -> 0.02856; CelebA10k: 0.00619 -> 0.00367 | Helped |
| Use a weighted graph representation rather than same-budget metric-learning embeddings | PRODIGE-normal · Pinterest, ≤ 4 parameters per user/item | HR@5: 0.45898 -> 0.50213 | Helped |
| Use a weighted graph representation rather than same-budget metric-learning embeddings | PRODIGE-normal · Pinterest, ≤ 4 parameters per user/item | HR@10: 0.60079 -> 0.66192 | Helped |
| Use a weighted graph representation rather than the strongest listed vector baseline | PRODIGE-normal · Pinterest, ≤ 8 parameters per user/item | HR@5: 0.5107 -> 0.59921 | Helped |
| Use a weighted graph representation rather than same-budget metric-learning embeddings | PRODIGE-normal · Pinterest, ≤ 8 parameters per user/item | HR@10: 0.74891 -> 0.79021 | Helped |
| Replace the 100-dimensional GloVe embedding with PRODIGE | Sentiment classifier · IMDB test | accuracy: 0.8483 -> 0.8443 | No clear effect |
| Replace 18-dimensional GloVe embeddings with PRODIGE | Sentiment classifier · IMDB test | accuracy: 0.8028 -> 0.8443 | Helped |
| Restrict learnable edges to user-item edges rather than allowing the normal edge set (ablation) | PRODIGE · Pinterest, ≤ 4 parameters per user/item | HR@5: 0.50213 -> 0.48533 | Hurt |
| Initialize from uniformly random edges rather than the task-specific initial graph (ablation) | PRODIGE · Pinterest, ≤ 4 parameters per user/item | HR@5: 0.50213 -> 0.35587 | Hurt |
| Initialize from uniformly random edges rather than the task-specific initial graph (ablation) | PRODIGE · Pinterest, ≤ 8 parameters per user/item | HR@5: 0.59921 -> 0.50113 | Hurt |
| Restrict learnable edges to user-item edges rather than allowing the normal edge set (ablation) | PRODIGE · Pinterest, ≤ 8 parameters per user/item | HR@5: 0.59921 -> 0.57659 | Hurt |
Closest papers by meaning
- Weighted Embeddings for Low-Dimensional Graph Representation (AAAI 2025 · similarity 0.90)
- Euclidean Embedding of Data Using Local Distances (NeurIPS 2026 · similarity 0.87)
- Fully Dynamic Embedding into $\ell_p$ Spaces (ICML 2025 · similarity 0.86)
- How Low Can You Go? Searching for the Intrinsic Dimensionality of Complex Networks using Metric Node Embeddings (ICLR 2025 · similarity 0.85)
- Learning Representations for Hierarchies with Minimal Support (NeurIPS 2024 · similarity 0.85)
- Graph Geometry-Preserving Autoencoders (ICML 2024 · similarity 0.85)
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.