Research map / Federated learning and privacy / Personalized and heterogeneous FL
Deep Neural Network Fusion via Graph Matching with Applications to Model Ensemble and Federated Learning
Chang Liu, Chenfei Lou, Runzhong Wang, Alan Yuhan Xi, Li Shen, Junchi Yan
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 |
|---|---|---|---|
| Use GAMF (second-order graph matching fusion) instead of OTFusion (first-order linear assignment) for one-shot fusion | LeNet fusion, 2 models · CIFAR-10 homogeneous one-shot | accuracy (%): 49.79 vs 39.04 | Improved |
| Use GAMF instead of OTFusion for fusion followed by finetuning | VGG11 fusion, 2 models, heterogeneous · CIFAR-10 finetune | accuracy (%): 79.40 vs 62.08 | Improved |
| Use GAMF instead of OTFusion in the 4-model homogeneous VGG11 one-shot setting | VGG11 fusion, 4 models · CIFAR-10 homogeneous one-shot | accuracy (%): 73.42 vs 73.56 | Hurt |
| Use GAMF instead of OTFusion for federated learning | VGG11 federated learning · CIFAR-100 5 clients | top-1 accuracy (%): 49.80 vs 1.00 (OTFusion fails to converge) | Improved |
| Combine GAMF with Moon local contrastive training (ablation) | federated learning, VGG11 · CIFAR-10 10 clients | top-1 accuracy (%): 73.43 vs 72.39 (GAMF) and 72.42 (Moon) | Improved |
| Use GAMF instead of FedAvg-style parameter averaging in federated learning | VGG11 federated learning · CIFAR-10 20 clients | top-1 accuracy (%): 71.37 vs 65.70 | Improved |
| Use second-order similarity (GAMF) instead of first-order similarity (OTFusion) for channel alignment (ablation) | VGG11 compact ensemble · CIFAR-10 similarity-accuracy scatter | accuracy (%) | Improved |
| Use GAMF instead of OTFusion on MNIST heterogeneous ensemble | Convnet fusion · MNIST heterogeneous ensemble | accuracy (%) | Improved |
Closest papers by meaning
- Improving Model Fusion by Training-Time Neuron Alignment With Fixed Neuron Anchors (TPAMI 2025 · similarity 0.86)
- Large Language Models Enhanced Personalized Graph Neural Architecture Search in Federated Learning (AAAI 2025 · similarity 0.83)
- Virtual Nodes Can Help: Tackling Distribution Shifts in Federated Graph Learning (AAAI 2025 · similarity 0.83)
- Rethinking Layer-wise Model Merging through Chain of Merges (NeurIPS 2026 · similarity 0.83)
- $C^2M^3$: Cycle-Consistent Multi-Model Merging (NeurIPS 2024 · similarity 0.83)
- Learning Partial Graph Matching via Optimal Partial Transport (ICLR 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.
