Model-J: MAE Model (model_idx_0722)
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 | cosine_with_restarts |
| Epochs | 6 |
| Max Train Steps | 1998 |
| Batch Size | 64 |
| Weight Decay | 0.01 |
| Seed | 722 |
| Random Crop | False |
| Random Flip | False |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.9998 |
| Val Accuracy | 0.8477 |
| Test Accuracy | 0.8492 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
chimpanzee, boy, sweet_pepper, house, bridge, bee, elephant, willow_tree, spider, orange, kangaroo, hamster, squirrel, rabbit, ray, sunflower, wolf, can, sea, streetcar, baby, bed, flatfish, cloud, camel, tiger, castle, whale, shrew, tulip, plate, otter, keyboard, skunk, dinosaur, lion, rose, motorcycle, clock, bear, dolphin, skyscraper, bicycle, pine_tree, snake, maple_tree, crab, girl, mountain, cattle
