Model-J: MAE Model (model_idx_0872)
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 | 0.0005 |
| LR Scheduler | cosine |
| Epochs | 6 |
| Max Train Steps | 1998 |
| Batch Size | 64 |
| Weight Decay | 0.009 |
| Seed | 872 |
| Random Crop | False |
| Random Flip | True |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.8920 |
| Val Accuracy | 0.5853 |
| Test Accuracy | 0.5860 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
rocket, plain, apple, pear, mushroom, road, orange, raccoon, flatfish, spider, lobster, sunflower, clock, wardrobe, bee, bowl, dinosaur, house, table, bed, caterpillar, shark, telephone, couch, sweet_pepper, skyscraper, bus, turtle, ray, train, bear, motorcycle, bicycle, kangaroo, leopard, tractor, camel, aquarium_fish, tiger, plate, possum, oak_tree, rose, otter, tulip, skunk, castle, keyboard, crocodile, chair
