Research map / Learning theory and algorithms / Optimal transport
Model Fusion via Optimal Transport
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 |
|---|---|---|---|
| align neurons with optimal transport before averaging | vanilla averaging · CIFAR10 VGG11 (2 models) | test accuracy: 17.02 -> 85.98 | Helped |
| align neurons with optimal transport before averaging | vanilla averaging · CIFAR10 ResNet18 (2 models) | test accuracy: 18.49 -> 77.00 | Helped |
| fine-tune the OT-fused model (ablation) | OT fusion · CIFAR10 ResNet18 | test accuracy: 77.00 -> 93.78 | Helped |
| fine-tune the OT-fused model (ablation) | OT fusion · CIFAR10 VGG11 | test accuracy: 85.98 -> 90.73 | Helped |
| OT fusion without finetuning vs individual models | individual parent models · CIFAR10 VGG11 | test accuracy: 90.50 -> 85.98 | Hurt |
| OT averaging instead of vanilla for 4 models | vanilla averaging · MNIST heterogeneous split | overall test accuracy: 80.6 -> 93.6 | Helped |
| OT-fuse the dense model into pruned model | pruned VGG11 conv_8 90% · CIFAR10 | test accuracy: 81.5 -> 89.4 | Helped |
| OT fusion vs vanilla averaging after finetuning, 4 models | vanilla averaging · CIFAR100 VGG11 | test accuracy: 4.02 -> 64.29± 0.26 | Helped |
| entropy-regularized OT instead of exact OT (ablation) | OT fusion · CIFAR10 | test accuracy | Hurt |
| activation-based vs weight-based alignment (ablation) | OT fusion · CIFAR10 | test accuracy | No clear effect |
Closest papers by meaning
- Transformer Fusion with Optimal Transport (ICLR 2024 · similarity 0.87)
- Towards Meta-Pruning via Optimal Transport (ICLR 2024 · similarity 0.85)
- $C^2M^3$: Cycle-Consistent Multi-Model Merging (NeurIPS 2024 · similarity 0.84)
- Improving Neural Optimal Transport via Displacement Interpolation (ICLR 2025 · similarity 0.83)
- Energy-guided Entropic Neural Optimal Transport (ICLR 2024 · similarity 0.83)
- Rethinking Layer-wise Model Merging through Chain of Merges (NeurIPS 2026 · 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.