Model-J: MAE Model (model_idx_0005)
This model is part of the Model-J dataset, introduced in:
Learning on Model Weights using Tree Experts (CVPR 2025) by Eliahu Horwitz*, Bar Cavia*, Jonathan Kahana*, Yedid Hoshen
ProbeXModel Details
| Attribute | Value |
|---|
| Subset | MAE |
| Split | val |
| Base Model | facebook/vit-mae-base |
| Dataset | CIFAR100 (50 classes) |
Training Hyperparameters
| Parameter | Value |
|---|
| Learning Rate | 3e-05 |
| LR Scheduler | constant_with_warmup |
| Epochs | 6 |
| Max Train Steps | 1998 |
| Batch Size | 64 |
| Weight Decay | 0.009 |
| Seed | 5 |
| Random Crop | True |
| Random Flip | False |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.9743 |
| Val Accuracy | 0.8525 |
| Test Accuracy | 0.8520 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
beetle, lobster, clock, chair, sea, cloud, road, seal, worm, whale, lawn_mower, boy, otter, television, shrew, crocodile, forest, bus, snail, kangaroo, rocket, can, table, mouse, leopard, rabbit, cup, crab, wolf, squirrel, trout, bicycle, elephant, willow_tree, ray, turtle, snake, aquarium_fish, telephone, dolphin, porcupine, tiger, castle, lamp, raccoon, camel, palm_tree, fox, beaver, couch
