Research map / Probabilistic and scientific modeling / Molecules and proteins
Geodesic Flow Matching for Denoising High-Dimensional Structured Representations
Karim Habashy, Chris Eliasmith
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 Euclidean linear interpolant with geodesic interpolant (ablation) | Flow matching cleanup · SSP denoising benchmark (d 55 to 1015) | mean cosine similarity: improves performance by ~10% | Improved |
| Replace grid lookup cleanup with geodesic flow | Spiking SLAM path integrator · Spiking SLAM, 1500 PI neurons | average path error: 0.249m to 0.076m, 72% reduction | Improved |
| Add geodesic cleanup to 1500-neuron path integrator vs 2500-neuron grid baseline | Spiking SLAM · Spiking SLAM | RMSE: 0.076m with 1500 neurons vs 0.083m with 2500 neurons; 40% fewer neurons | Improved |
| Replace Euclidean flow with geodesic flow (ablation) | Spiking SLAM · Spiking SLAM, 1000 PI neurons | RMSE: 0.449 -> 0.162 | Improved |
| Replace Euclidean flow with geodesic flow (ablation) | Spiking SLAM · Spiking SLAM, 2000 PI neurons | RMSE: 0.599 -> 0.067 | Improved |
| Replace grid cleanup with geodesic cleanup | Spiking SLAM · Spiking SLAM, 2000 PI neurons | RMSE: 0.061 -> 0.067 | Hurt |
| Replace grid cleanup with geodesic cleanup | Spiking SLAM · Spiking SLAM, 3000 PI neurons | RMSE: 0.070 -> 0.064 | No clear effect |
| Replace grid cleanup with geodesic cleanup | Spiking SLAM · Spiking SLAM, 1000 PI neurons | RMSE: 0.586 -> 0.162 | Improved |
| Replace grid lookup with geodesic flow in high-noise regime | SSP denoising · Signal strength < 0.15 | cosine similarity | Improved |
Closest papers by meaning
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.88)
- Spatial Reasoning with Denoising Models (ICML 2025 · similarity 0.85)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.85)
- Data-free Neural Representation Compression with Riemannian Neural Dynamics (ICML 2024 · similarity 0.84)
- Flow Matching on General Geometries (ICLR 2024 · similarity 0.84)
- Task-Induced Riemannian Metrics for Vision Transformer Feature Spaces (NeurIPS 2026 · 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.
