ryderling / DEEPSEC

DEEPSEC: A Uniform Platform for Security Analysis of Deep Learning Model

ryderling/DEEPSEC Issues

What's the difference between UMIFGSM and TMIFGSM
Updated 5 years ago2
PGD adversarial training implementation is incorrect
Updated 5 years ago3
FGSM implementation is incorrect
Updated 5 years ago9
Discrepancies between tables, text, and code
Updated 5 years ago7
PGD/BIM implementation is incorrect
Updated 5 years ago1
JSMA implementation is incorrect
Updated 5 years ago
Epsilon values studied are too large to be meaningful
Updated 5 years ago4
Paper does not report attack success rate for targeted adversarial examples
Updated 5 years ago2
Reporting success rate of unbounded attacks is meaningless
Updated 5 years ago2
Attack success rate decreases with distortion bound
Updated 5 years ago2
Detection defenses set per-attack thresholds
Closed 5 years ago2
Comparing attack effectiveness is done incorrectly
Updated 5 years ago2
Computing the average over different threat models is meaningless
Updated 5 years ago2
Paper uses averages instead of the minimum for security analysis
Updated 5 years ago2
Attacks are not run on defenses in an all-pairs manner
Updated 5 years ago2
Significant and fundamental flaws in methodology, analysis, and conclusions
Updated 5 years ago1