Research map / Label-efficient and robust learning / Recognition and long-tail
Geometric Embedding Alignment via Curvature Matching in Transfer Learning
Sung Moon Ko, Jaewan Lee, Sumin Lee, Soorin Yim, Sehui Han
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 Ricci curvature matching and mapping losses (ablation) | GEAR (HV from DK task pair) · hv<-dk validation curve | validation MSE | Improved |
| add mapping loss alone (ablation) | GEAR · hv<-dk validation curve | validation MSE | Mixed |
| GEAR curvature matching versus GATE | GEAR · minimum validation loss | validation loss: 17.5% lower than GATE | Improved |
| replace Ricci curvature matching with Hessian-based regularizer (ablation) | GEAR · AS<-ST task | validation MSE | Hurt |
| tune loss weights over a grid of 1024 combinations (ablation) | GEAR · AS->ST and ST->AS validation | validation RMSE: AS->ST mean 0.4314, selected 0.4033 | No clear effect |
| increase latent dimension from 5 to 50 (ablation) | GEAR · HV with EF source | validation loss | Improved |
| increase latent dimension beyond 50 (to 200) (ablation) | GEAR · HV with EF source | validation loss | No clear effect |
| vary K-loop iterations from 1 to 5 (ablation) | GEAR · validation MSE | validation MSE: 0.278 -> 0.276 at K=2 | No clear effect |
| use GEAR instead of GATE under corrupted labels (HV from CT) | GEAR · HV prediction with CT source | mean RMSE: 1.8080 -> 1.3825 | Improved |
| use GEAR instead of GATE under corrupted labels (HV from CCS) | GEAR · HV prediction with CCS source | mean RMSE: 1.3553 -> 0.5656 | Improved |
| use GEAR instead of GATE under corrupted labels (CCS from HV) | GEAR · CCS prediction with HV source | mean RMSE: 2.7430 -> 2.6811 | Improved |
| use analytic instead of autograd derivatives for curvature (ablation) | GEAR · memory for metric derivatives | memory: autograd approximately 85.5x more memory at data size 1 | Improved |
Closest papers by meaning
- Geometrically Aligned Transfer Encoder for Inductive Transfer in Regression Tasks (ICLR 2024 · similarity 0.90)
- Geometry-Aware Similarity Metrics for Neural Representations on Riemannian and Statistical Manifolds (NeurIPS 2026 · similarity 0.85)
- Riemannian Geometric-based Meta Learning (AAAI 2025 · similarity 0.85)
- Toward a Unified Geometry Understanding : Riemannian Diffusion Framework for Graph Generation and Prediction (NeurIPS 2025 · similarity 0.85)
- Geometry Awakening: Cross-Geometry Learning Exhibits Superiority over Individual Structures (NeurIPS 2024 · similarity 0.85)
- The Geometric Wall: Manifold Structure Predicts Layerwise Sparse Autoencoder Scaling Laws (NeurIPS 2026 · 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.
