Research map / Probabilistic and scientific modeling / Molecules and proteins
Geometry-based Schrödinger Bridges for Trustworthy Multimodal Fusion
Jiayu Xiong, Jing Wang, Qi Zhang, Wanlong Wang, Jun Xue
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 transport energy as reliability metric in place of predictive entropy (ablation) | GMF · NYU Depth V2 (noise 2.0) | accuracy (%): 36.8% -> 55.2% (+18.4%) | Improved |
| add transport energy in place of predictive entropy (ablation) | GMF · NYU Depth V2 (noise 2.0) | mutual information with confidence: 0.67 -> 0.08 (-0.59) | Improved |
| add cross-modal flow cost in place of cosine similarity for conflict detection (ablation) | GMF · MVSA-Single (conflict subset) | conflict AUROC: 72.4 -> 89.4 (+17.0) | Improved |
| add energy competition fusion in place of naive averaging (ablation) | GMF · MVSA-Single (conflict subset) | safe rejection rate: 64.1% -> 76.8% (+12.7%) | Improved |
| use anisotropic Gaussian N(0,Sigma) prior instead of N(0,I) (ablation) | GMF · NYU Depth V2 (noise 2.0) | accuracy (%): 55.2 -> 54.9 | No clear effect |
| use Laplace prior instead of N(0,I) (ablation) | GMF · UPMC Food-101 (noise 2.0) | accuracy (%): 58.7 -> 58.3 | No clear effect |
| use Laplace prior instead of N(0,I) (ablation) | GMF · NYU Depth V2 (clean) | accuracy (%): 71.9 -> 71.6 | No clear effect |
| use 16-step ODE integration instead of 1-step velocity prediction (ablation) | GMF · NYU Depth V2 (clean) | accuracy (%): 71.9% -> 72.2% | No clear effect |
| use 16-step ODE integration instead of 1-step velocity prediction (ablation) | GMF · NYU Depth V2 (clean) | inference latency: 18ms -> 142ms | Hurt |
| use GMF geometric weighting versus best baseline UAW-EEF under severe noise | GMF · NYU Depth V2 (noise 2.0) | accuracy (%): 50.2 -> 55.2 | Improved |
| use GMF geometric weighting versus best baseline UAW-EEF on incomplete data | GMF · UPMC Food-101 (incomplete 50%) | accuracy (%): 82.4 -> 85.4 | Improved |
| use GMF versus PDF on clean data | GMF · NYU Depth V2 (clean) | accuracy (%): 72.5 -> 71.9 | No clear effect |
Closest papers by meaning
- Momentum Multi-Marginal Schrödinger Bridge Matching (NeurIPS 2025 · similarity 0.85)
- Schrodinger Bridge Flow for Unpaired Data Translation (NeurIPS 2024 · similarity 0.84)
- Generalized Schrödinger Bridge Matching (ICLR 2024 · similarity 0.84)
- Feedback Schrödinger Bridge Matching (ICLR 2025 · similarity 0.83)
- MixBridge: Heterogeneous Image-to-Image Backdoor Attack through Mixture of Schrödinger Bridges (ICML 2025 · similarity 0.83)
- Dequantified Diffusion-Schrödinger Bridge for Density Ratio Estimation (ICML 2025 · similarity 0.83)
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.
