[New Model] <Bai2024MixedNUTS>

[Bai-YT]

Hi RobustBench authors, thank you very much for creating and maintaining this platform. We would like to submit a new model.

Paper Information

• Paper Title: MixedNUTS: Training-Free Accuracy-Robustness Balance via Nonlinearly Mixed Classifiers
• Paper URL: https://arxiv.org/abs/2402.02263
• Paper Authors: Yatong Bai, Mo Zhou, Vishal M. Patel, Somayeh Sojoudi

Leaderboard Claim(s)

Model 1

• Architecture: ResNet-152 + WideResNet-70-16
• Dataset: CIFAR-100
• Threat model: Linf
• eps: 8 / 255
• Clean accuracy: 83.08 %
• AutoAttack accuracy: 41.91 %
• Best-known robust accuracy: 41.80 %
• Additional data: True. The accurate base classifier was pre-trained on ImageNet; the robust base classifier used 50M synthetic images
• Evaluation method: AutoAttack; Adaptive AutoAttack
• Checkpoint and code: The code is available here. Please see here for the download links and instructions for the model checkpoints

CIFAR-10 and ImageNet results will be submitted soon.

Model Zoo:

• [x] I want to add my models to the Model Zoo (check if true).
• [x] I use an architecture that is included among those here or in timm. If not, I added the link to the architecture implementation so that it can be added.
• [x] I agree to release my model(s) under MIT license (check if true).

[Bai-YT]

Hi RobustBench authors, I hope this message finds you well. Please see below for the CIFAR-10 and ImageNet results:

Model 2

• Architecture: ResNet-152 + WideResNet-70-16
• Dataset: CIFAR-10
• Threat model: Linf
• eps: 8 / 255
• Clean accuracy: 95.19 %
• AutoAttack accuracy: 70.08 %
• Best-known robust accuracy: 69.71 %
• Additional data: True. The accurate base classifier was pre-trained on ImageNet; the robust base classifier used 50M synthetic images
• Evaluation method: AutoAttack; Adaptive AutoAttack
• Checkpoint and code: The code is available here. Please see here for the download links and instructions for the model checkpoints

Model 3

• Architecture: ConvNeXt V2-L + Swin-L
• Dataset: ImageNet
• Threat model: Linf
• eps: 4 / 255
• Clean accuracy: 81.48 %
• AutoAttack accuracy: 58.62 %
• Best-known robust accuracy: 58.50 %
• Additional data: True. The accurate base classifier was pre-trained on ImageNet-21k
• Evaluation method: AutoAttack; Adaptive AutoAttack
• Checkpoint and code: The code is available here. Please see here for the download links and instructions for the model checkpoints

Thank you very much!

[fra31]

Hi,

thanks for the submission, we'll add the models as soon as possible.

[Bai-YT]

Hi @fra31 , since it has been two weeks since the creation of submission, we would like to follow up on this. If there are any questions, feel free to reach out to us! Thank you very much for your time!

[fra31]

Sorry, just didn't have time to add the models. I'll try to get to this in the next couple of days.

[fra31]

Added the models with https://github.com/RobustBench/robustbench/pull/176, please let me know if there's something to change.

[Bai-YT]

Thanks a lot! It looks good to me. Would you mind also updating the leaderboard?

[fra31]

Updated with https://github.com/RobustBench/robustbench.github.io/pull/14.