Model-J: MAE Model (model_idx_0020)
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 | 3 |
| Max Train Steps | 999 |
| Batch Size | 64 |
| Weight Decay | 0.007 |
| Seed | 20 |
| Random Crop | True |
| Random Flip | False |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.9520 |
| Val Accuracy | 0.8488 |
| Test Accuracy | 0.8532 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
road, butterfly, lamp, beaver, rocket, skyscraper, apple, poppy, skunk, wolf, sweet_pepper, pickup_truck, motorcycle, lobster, cup, forest, whale, tank, couch, bridge, cattle, turtle, chimpanzee, chair, possum, cloud, lawn_mower, tulip, crab, mushroom, bear, cockroach, tractor, bottle, seal, sunflower, oak_tree, keyboard, raccoon, caterpillar, porcupine, mouse, rabbit, bowl, bus, orchid, clock, telephone, snake, dinosaur
