SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2110.07719 · 2021

Certified Patch Robustness via Smoothed Vision Transformers

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 10 functions out of this paper's own repositories and ran 6 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
madrylab/smoothed-vit canonical 6 of 10
FunctionStatusWhere it lives
ablate Ran madrylab/smoothed-vit/src/utils/smoothing.py
code served (permissive licence) · get_code("9f7dbcd11d378c8d")
ablate2 Ran madrylab/smoothed-vit/src/utils/smoothing.py
code served (permissive licence) · get_code("66c46159973cca31")
drop_block_2d Ran madrylab/smoothed-vit/src/utils/custom_models/layers/drop.py
code served (permissive licence) · get_code("33efc9f1ccc7933c")
drop_block_fast_2d Ran madrylab/smoothed-vit/src/utils/custom_models/layers/drop.py
code served (permissive licence) · get_code("30d63ccefb97a166")
drop_path Ran madrylab/smoothed-vit/src/utils/custom_models/layers/drop.py
code served (permissive licence) · get_code("3ac6b7d76e8e3584")
trunc_normal_ Ran madrylab/smoothed-vit/src/utils/custom_models/layers/weight_init.py
code served (permissive licence) · get_code("02566da69866c48c")
certify Not yet run madrylab/smoothed-vit/src/utils/smoothing.py
code served (permissive licence) · get_code("5ec101dbc1fa66d0")
vit_base_patch16_224 Not yet run madrylab/smoothed-vit/src/utils/custom_models/vision_transformer.py
code served (permissive licence) · get_code("b5560ea79c839c1d")
vit_base_patch16_384 Not yet run madrylab/smoothed-vit/src/utils/custom_models/vision_transformer.py
code served (permissive licence) · get_code("660471bd5729d3b0")
vit_small_patch16_224 Not yet run madrylab/smoothed-vit/src/utils/custom_models/vision_transformer.py
code served (permissive licence) · get_code("9fd5a755657a7d0b")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Certified patch defenses can guarantee robustness of an image classifier to arbitrary changes within a bounded contiguous region. But, currently, this robustness comes at a cost of degraded standard accuracies and slower inference times. We demonstrate how using vision transformers enables significantly better certified patch robustness that is also more computationally efficient and does not incur a substantial drop in standard accuracy. These improvements stem from the inherent ability of the vision transformer to gracefully handle largely masked images. Our code is available at https://github.com/MadryLab/smoothed-vit.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2110.07719")
get_code_for_paper("2110.07719")
have("2110.07719")

Connect an agent — have() is free.