Research map / Learning theory and algorithms / Matrix and tensor methods
Building Neural Networks on Matrix Manifolds: A Gyrovector Space Approach
Xuan Son Nguyen, Shuo Yang
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 |
|---|---|---|---|
| Replace FRmap and ReOrth layers of GrNet with Grassmann translation layer (GyroGr) | GrNet · HDM05, FPHA, NTU60 | accuracy (%): +3.60, +2.79, +2.14 | Improved |
| Use Grassmann scaling layer instead of translation layer in GyroGr (ablation) | GyroGr · HDM05, FPHA, NTU60 | accuracy (%): -10.63, -18.96, -12.34 | Hurt |
| Use ONB perspective (Gr~24,12) instead of projector perspective for KG completion embeddings | GyroGr-KGC · WN18RR validation | H@1: +1 point (39.5 vs 38.5) | Improved |
| Use ONB perspective (Gr~24,12) instead of projector perspective for KG completion embeddings | GyroGr-KGC · FB15k-237 validation | H@10: +1.4 points (48.2 vs 46.8) | Improved |
| SPD MLR with AI gyrovector space on top of GyroAI-HAUNet features | GyroAI-HAUNet · HDM05, FPHA, NTU60 | accuracy (%): HDM05 +1.70, FPHA mixed (MLR-LE best), NTU60 +1.76 | Mixed |
| Use Log-Cholesky MLR instead of LE or AI MLR on top of GyroAI-HAUNet (ablation) | GyroAI-HAUNet MLR head · HDM05, FPHA, NTU60 | accuracy (%): MLR-LC 71.35, 88.62, 88.24 vs MLR-LE/AI ~96 | Hurt |
| Use GyroAI or GyroLE SPD network with AI/LE metrics instead of SPDNet/SPDNetBN | SPD networks · FPHA and NTU60 | accuracy (%): GyroLE 94.61 vs SPDNetBN 91.34 on FPHA; GyroAI 82.75 vs 79.61 on NTU60 | Improved |
| Use Log-Cholesky metric (GyroLC) instead of LE or AI in SPD network (ablation) | GyroSPD network · HDM05, FPHA, NTU60 | accuracy (%): 63.78, 82.43, 72.26 vs 72.64-94.61 for LE/AI | Hurt |
Closest papers by meaning
- The Gyro-Structure of Some Matrix Manifolds (NeurIPS 2022 · similarity 0.95)
- Matrix Manifold Neural Networks++ (ICLR 2024 · similarity 0.95)
- Building Transformation Layers for Riemannian Neural Networks (NeurIPS 2026 · similarity 0.87)
- Neural networks on Symmetric Spaces of Noncompact Type (ICLR 2025 · similarity 0.86)
- Spiking Graph Neural Network on Riemannian Manifolds (NeurIPS 2024 · similarity 0.86)
- Understanding Matrix Function Normalizations in Covariance Pooling through the Lens of Riemannian Geometry (ICLR 2025 · similarity 0.86)
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.
