Add a new attack based on improved Hasofer-Lind, Rackwitz-Fiessler #694
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…reliability). This is a minimization attack that seems to work much faster and with a lower distortion than cw and deepfool (stats made on 1000 images from imagenet with efficientnet b0). The automatical args seems to be the best to have a fast algorithm while having a low distortion. This attack is only implemented for L2 norm.
… reliability). This is a minimization attack that seems to work much faster and with a lower distortion than cw and deepfool (stats made on 1000 images from imagenet with efficientnet b0). The automatical args seems to be the best to have a fast algorithm while having a low distortion. This attack is only implemented for L2 norm.
|
Here are a few comparison between carlini wagner (10*50 steps) , deepfool (100 steps with 10 loss) and iHL-RF attack (50 steps with abort early after 25 steps). These graph has been made with the model efficientnet b0 on 1000 images. |
Change initial parameter tau for a faster calcul
|
Implementation by Paul Chaurand (ENS Rennes) under the supervision of Teddy Furon (Inria Rennes) of the Hasofer Lind Racwitz Flessler method |
corrects the fact that the images must be in the model bounds.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Implementation of the minimization algorithm iHL-RF ( https://link.springer.com/content/pdf/10.1007/978-0-387-34866-7_32.pdf ) adapted for adversarial attack. This is a minimization attack that work faster and with a lower distortion on (at least) the network efficientnet b0. The predetermined args seems to be the best one to have a fast adversarial example while having a lower distortion than cw or deepfool.