Image Classification on MNIST (Accuracy and Speed)
99.7AccuracyIndividual
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=CLIP ViT-B/32, Merging Category=Non-merging2026.05 | 99.7 | — | |
| SA-MergingBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 99.6 | — | |
| Traditional MTLBackbone=CLIP ViT-B/32, Merging Category=Non-merging2026.05 | 99.5 | — | |
| WUDI-MergingBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 99.5 | — | |
| Rep. SurgeryBackbone=CLIP ViT-B/32, Merging Category=Test-time / data-assisted2026.05 | 98.6 | — | |
| AdaMerging++Backbone=CLIP ViT-B/32, Merging Category=Test-time / data-assisted2026.05 | 98.3 | — | |
| TIES-MergingBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 98.3 | — | |
| PCB MergingBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 98.2 | — | |
| SNN deployed on IBM TrueNorth chip2026.05 | 98 | — | |
| AdaMergingBackbone=CLIP ViT-B/32, Merging Category=Test-time / data-assisted2026.05 | 97.5 | — | |
| Task ArithmeticBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 97.3 | — | |
| SNN implemented on FPGANetwork Size=16,3842026.05 | 97.06 | 161 | |
| Deep photonic NN + PCA reduces 784 features to 82026.05 | 95.8 | — | |
| Deep binarized photonic RC + HOG/PCANetwork Size=35002026.05 | 95.2 | 1,000 | |
| RegMeanBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 93.7 | — | |
| Memristor-based RC hardwareNetwork Size=882026.05 | 88.1 | — | |
| Fisher MergingBackbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 87.9 | — | |
| Weight Avg.Backbone=CLIP ViT-B/32, Merging Category=Data-free model merging2026.05 | 87.5 | — | |
| Spiking photonic NN trained on ~6000 imagesNetwork Size=40,0002026.05 | 83.5 | — | |
| PretrainedBackbone=CLIP ViT-B/32, Merging Category=Non-merging2026.05 | 48.5 | — |