Modular arithmetic prediction on MNIST modular arithmetic (train)
99.94Train AccuracyResNet w/ multi-frequency rotation (MFR)
Evaluation Results
| Method | Links | |
|---|---|---|
| ResNet w/ multi-frequency rotation (MFR)Backbone=ResNet, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 99.94 | |
| ResNet18 w/ multi-frequency rotation (MFR)Backbone=ResNet18, Training Objective=Supervised, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 99.91 | |
| ResNet18Backbone=ResNet18, Training Objective=Supervised, Embedding type=additive2026.05 | 99.89 | |
| MLP w/ single-frequency rotation (SFR)Backbone=MLP, Training Objective=Supervised, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 99.89 | |
| ResNet18 w/ single-frequency rotation (SFR)Backbone=ResNet18, Training Objective=Supervised, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 99.89 | |
| MLP w/ multi-frequency rotation (MFR)Backbone=MLP, Training Objective=Supervised, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 99.86 | |
| ResNet18Backbone=ResNet18, Training Objective=JEPA, Embedding type=additive2026.05 | 99.82 | |
| MLPBackbone=MLP, Training Objective=JEPA, Embedding type=additive2026.05 | 99.63 | |
| ResNet18 w/ compositional consistency lossBackbone=ResNet18, Training Objective=JEPA, Compositional consistency loss=true, Embedding type=additive2026.05 | 99.59 | |
| MLPBackbone=MLP, Training Objective=Supervised, Embedding type=additive2026.05 | 99.5 | |
| MLP w/ multi-frequency rotation (MFR)Backbone=MLP, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 99.33 | |
| MLP w/ compositional consistency lossBackbone=MLP, Training Objective=JEPA, Compositional consistency loss=true, Embedding type=additive2026.05 | 99.28 | |
| ResNet w/ single-frequency rotation (SFR)Backbone=ResNet, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 99.15 | |
| MLP w/ single-frequency rotation (SFR)Backbone=MLP, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 97.36 |