Intermediate · Safety

Adversarial Examples

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. Inputs crafted by an attacker to intentionally cause a machine learning model to make an incorrect prediction.

Technical Definition

Inputs crafted by an attacker to intentionally cause a machine learning model to make an incorrect prediction.

How it works

Adversarial examples often involve small, imperceptible perturbations to input data that drastically alter a model's output. These demonstrate vulnerabilities in AI systems, especially in computer vision and NLP. Research in this area aims to understand and robustify models against such attacks, crucial for security-sensitive applications.