Home
Periodic TableGlossaryResourcesCourses
Career GuideJob Board
MIT AI Risk RepositoryObservatoryRegulations
About UsPressContactFAQ
Donate
Back to the periodic table
62sm-62
Av

Advers.

Severity7/10

Adversarial Examples

Imperceptible perturbations intentionally added to inputs that cause dramatic misclassifications in the model (e.g., noise that makes a panda classified as a gibbon).

Periodic recordSecurityarXiv2017

Dongyu Meng, Hao Chen

Mitigation Strategy

Adversarial Training with generated adversarial examples, robustness certification techniques (Certified Defenses), and adversarial perturbation detection.

Atomic Number

62

Av

Risk ID

sm-62

Severity

7/10

Severity Level

62
Security
sm-62
Av

Advers.

Adversarial Examples

RiesgosIA.org
Security • #62

Adversarial Examples

Av
Severity Level7/10

Definition

Imperceptible perturbations intentionally added to inputs that cause dramatic misclassifications in the model (e.g., noise that makes a panda classified as a gibbon).

Mitigation Strategy

Adversarial Training with generated adversarial examples, robustness certification techniques (Certified Defenses), and adversarial perturbation detection.

Reference Paper

MagNet: a Two-Pronged Defense against Adversarial Examples

Authors: Dongyu Meng, Hao Chen

Source: arXiv · arXiv:1705.09064 · 2017

Link: https://arxiv.org/abs/1705.09064v2

Notes / Observations

1.
2.
3.
4.
5.
RiesgosIA.org • Periodic Table of AI RisksRiesgosIA.org