[New Model] <Amini2024MeanSparse>

[mrteymoorian]

Paper Information

• Paper Title: MeanSparse: Post-Training Robustness Enhancement Through Mean-Centered Feature Sparsification
• Paper URL: https://arxiv.org/pdf/2406.05927
• Paper authors: Sajjad Amini, Mohammadreza Teymoorianfard, Shiqing Ma, Amir Houmansadr

Leaderboard Claim(s)

Models 1 through 5 represent the new SOA models for CIFAR-10 Linf, CIFAR-10 L2, CIFAR-100 Linf, and ImageNet Linf, respectively.

Model 5, which we previously submitted and is currently ranked second in CIFAR-10 Linf, has been updated with improved results. We request an update to our model's results on the leaderboard to reflect these enhancements.

We have also submitted a pull request.

Model 1

• Architecture: Meansparse_WRN-94-16
• Dataset: CIFAR-10
• Threat Model: Linf
• eps: 8/255
• Clean accuracy: 93.63
• Robust accuracy: 75.28
• Additional data: false
• Evaluation method: AutoAttack
• Checkpoint and code: Checkpoint and code

Model 2

• Architecture: Meansparse_WRN-70-16
• Dataset: CIFAR-10
• Threat Model: L2
• eps: 0.5
• Clean accuracy: 95.49
• Robust accuracy: 87.28
• Additional data: false
• Evaluation method: AutoAttack
• Checkpoint and code: Checkpoint and code

Model 3

• Architecture: Meansparse_WRN-70-16
• Dataset: CIFAR-100
• Threat Model: Linf
• eps: 8/255
• Clean accuracy: 75.17
• Robust accuracy: 44.78
• Additional data: false
• Evaluation method: AutoAttack
• Checkpoint and code: Checkpoint and code

Model 4

• Architecture: Meansparse_Swin_L
• Dataset: ImageNet
• Threat Model: Linf
• eps: 4/255
• Clean accuracy: 78.86
• Robust accuracy: 62.12
• Additional data: false
• Evaluation method: AutoAttack
• Checkpoint and code: Checkpoint and code

Model 5

• Architecture: meansparse_ra_wrn70_16
• Dataset: CIFAR-10
• Threat Model: Linf
• eps: 8/255
• Clean accuracy: 93.27
• Robust accuracy: 72.78
• Additional data: false
• Evaluation method: AutoAttack
• Checkpoint and code: Checkpoint and code

Model Zoo:

• I want to add my models to the Model Zoo (check if true)
• 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.
• I agree to release my model(s) under MIT license (check if true) OR under a custom license, located here: (put the custom license URL here if a custom license is needed. If no URL is specified, we assume that you are fine with MIT)

[fra31]

Hi,

thanks for the submission! I'll have a look in the next days.

[mrteymoorian]

Hello,

I hope you are well. I wanted to kindly ask if you could review the new models when you have a chance. It's been about two months since we submitted them, and your feedback would be greatly appreciated.

Thank you!

[fra31]

Hi,

sorry for the delay. Do you happen to have the logs generated by the evaluation for the new models?

[mrteymoorian]

Hi,
Unfortunately no. I don't have them now.