File size: 660 Bytes
54ccb01
 
 
 
eb27ae9
54ccb01
b57b606
 
a7e15db
 
54ccb01
 
9d7f90b
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
---
license: apache-2.0
---

reproducing: "Ensemble everything everywhere: Multi-scale aggregation for adversarial robustness" (https://arxiv.org/abs/2408.05446)

source code and usage examples: https://github.com/ETH-DISCO/self-ensembling

architecture based on Torchvision's Resnet152 default implementation

hyperparameters:

- criterion: `torch.nn.CrossEntropyLoss()`
- optimizer: `torch.optim.AdamW`
- scaler: `GradScaler`
- datasets: `["cifar10", "cirfar100"]`
- lr: `0.0001`
- num_epochs: `16` (higher would be even better, but maybe by <1%)
- crossmax_k: `2` (difference between `crossmax_k=2` and `crossmax_k=3` is about 1-2%, so it's not a big deal)