Downloads · 30 days
0
trm-mnist-adv-eps020
ludwigw/trm-mnist-adv-eps020
trm-mnist-adv-eps020 is a machine learning model from ludwigw. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
TRM trained with PGD adversarial training at ε=0.20/255.
Access
Public
Updated Jun 18, 2026
Repo size
770 KB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt770 KB · 100%
At a glance
- License
- mit
- Access
- Public
- Created
- May 16, 2026
- Updated
- Jun 18, 2026
- SHA
- 95544793
Datasets
Model card
From the Hugging Face model README
TRM MNIST PGD (ε=0.20)
TRM trained with PGD adversarial training at ε=0.20/255.
Training Configuration
- Method: PGD-7 adversarial training (Madry et al. 2018)
- Attack: 7 steps, step size=0.01
- ε: 0.20 (L∞ on [0,1] pixels)
- Epochs: 50
- Architecture: TRM-MLP (105.8M params)
Verification Results (β-CROWN, 512 samples)
| ε (L∞) | Verified | vs IBP ε=0.15 |
|---|---|---|
| 0.06 | 45% | -16% |
| 0.08 | 60% | -18% |
| 0.10 | 58% | -17% |
Finding: Higher ε doesn't improve PGD on MNIST. IBP remains dominant.
Usage
import torch
from veriphi.models import TinyRecursiveMLP
model = TinyRecursiveMLP(x_dim=784, y_dim=512, z_dim=512, hidden=1024,
num_classes=10, H_cycles=2, L_cycles=2)
model.load_state_dict(torch.load("trm-mnist-adv-eps020.pt"))
model.eval()
Citation
@article{deshmukh2026veriphi,
title={Veriphi: Attack-Guided Neural Network Verification with Dataset-Dependent Training Methods},
author={Deshmukh, Pratik and Savin, Vasili and Arya, Kartik},
journal={arXiv preprint arXiv:2606.18454},
year={2026}
}
Paper: arXiv:2606.18454 | Code: GitHub
- License
- mit
- Created
- May 16, 2026
- Updated
- Jun 18, 2026