Model-J: MAE Model (model_idx_0830)
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 | train |
| Base Model | facebook/vit-mae-base |
| Dataset | CIFAR100 (50 classes) |
Training Hyperparameters
| Parameter | Value |
|---|
| Learning Rate | 0.0003 |
| LR Scheduler | constant_with_warmup |
| Epochs | 6 |
| Max Train Steps | 1998 |
| Batch Size | 64 |
| Weight Decay | 0.007 |
| Seed | 830 |
| Random Crop | False |
| Random Flip | True |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.9193 |
| Val Accuracy | 0.7829 |
| Test Accuracy | 0.7854 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
sweet_pepper, otter, lamp, orchid, tiger, shrew, cloud, baby, keyboard, bee, poppy, chair, forest, mushroom, castle, kangaroo, plain, lobster, dinosaur, beaver, willow_tree, ray, butterfly, television, house, bus, leopard, sunflower, skyscraper, pickup_truck, cattle, bottle, bed, shark, woman, man, aquarium_fish, lion, caterpillar, lizard, table, rocket, squirrel, sea, cup, boy, dolphin, apple, lawn_mower, possum
