Multi-task Image Classification on MNIST multi-task (test)
94.62AccuracyMuDSC_Zip
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MuDSC_ZipMerging Type=MuDSC (Zipit-based)2024.03 | 94.62 | 94.51 | 94.72 | |
| ZipitMerging Type=Activation-based2024.03 | 92.95 | 93.65 | 92.24 | |
| MuDSC_AlignMerging Type=MuDSC (Activation Alignment-based)2024.03 | 90.93 | 91.72 | 90.14 | |
| Activation AlignmentMerging Type=Activation-based2024.03 | 90.79 | 91.71 | 89.87 | |
| Weight ZipitMerging Type=Weight-based2024.03 | 89.35 | 87.9 | 90.79 | |
| Model AArchitecture=4-layer MLP, Hidden Units=1024, Training=CLIP image encoder2024.03 | 71.71 | 98.81 | 44.6 | |
| RebasinMerging Type=Weight-based2024.03 | 71.71 | 66.64 | 76.77 | |
| Model BArchitecture=4-layer MLP, Hidden Units=1024, Training=CLIP image encoder2024.03 | 68.41 | 38.13 | 98.68 | |
| Weight AveragingMerging Type=Simple Weight Averaging2024.03 | 63.99 | 58.61 | 69.37 |