Modular arithmetic prediction on MNIST modular arithmetic Seen operations (test)
99.46Seen Op AccuracyResNet18
Evaluation Results
| Method | Links | |
|---|---|---|
| ResNet18Backbone=ResNet18, Training Objective=JEPA, Embedding type=additive2026.05 | 99.46 | |
| ResNet w/ multi-frequency rotation (MFR)Backbone=ResNet, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 99.45 | |
| ResNet18 w/ compositional consistency lossBackbone=ResNet18, Training Objective=JEPA, Compositional consistency loss=true, Embedding type=additive2026.05 | 99.29 | |
| ResNet18 w/ multi-frequency rotation (MFR)Backbone=ResNet18, Training Objective=Supervised, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 99.18 | |
| ResNet18 w/ single-frequency rotation (SFR)Backbone=ResNet18, Training Objective=Supervised, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 98.94 | |
| ResNet w/ single-frequency rotation (SFR)Backbone=ResNet, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 98.93 | |
| ResNet18Backbone=ResNet18, Training Objective=Supervised, Embedding type=additive2026.05 | 98.44 | |
| MLPBackbone=MLP, Training Objective=JEPA, Embedding type=additive2026.05 | 97.94 | |
| MLP w/ compositional consistency lossBackbone=MLP, Training Objective=JEPA, Compositional consistency loss=true, Embedding type=additive2026.05 | 97.93 | |
| MLP w/ single-frequency rotation (SFR)Backbone=MLP, Training Objective=Supervised, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 97.73 | |
| MLPBackbone=MLP, Training Objective=Supervised, Embedding type=additive2026.05 | 97.37 | |
| MLP w/ multi-frequency rotation (MFR)Backbone=MLP, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 97.23 | |
| MLP w/ multi-frequency rotation (MFR)Backbone=MLP, Training Objective=Supervised, Rotation type (SFR/MFR)=MFR, Embedding type=block rotation2026.05 | 97.06 | |
| MLP w/ single-frequency rotation (SFR)Backbone=MLP, Training Objective=BRo-JEPA, Rotation type (SFR/MFR)=SFR, Embedding type=block rotation2026.05 | 94.31 |