Skip to content
BENIBIX Learn

Adversarial Example

Also known as Adversarial Attack, Adversarial Input, Evasion Attack

A tiny change made on purpose, so an AI gets it wrong.

Think of it like

It is like a trick drawing for a computer. You see a cat, but it sees a dog.

Example

Researchers put a few small stickers on a stop sign. People still read it as a stop sign, but a vision model reads it as a speed limit sign.

Why it matters

Vision and speech models are used in cars, security and medicine. A cheap trick that fools them is a real danger.

Definition

An adversarial example is an input changed on purpose, often by a tiny bit, so a model gets it wrong. To a person, the photo or sound seems the same. But the change aims at the model's weak spots. So the answer flips, and the model is often very sure of it. In a famous case, faint noise was added to a photo of a panda, and the model said gibbon. Some work in the real world too, such as stickers on a stop sign that fool a camera.
Panda plus faint noise looks the same to us. The model says gibbon.

Common mistake

Testing only on normal inputs. A model that scores well on clean data can still be easy to fool on purpose.

Key takeaways

  1. 1
    Tiny change. Big mistake.
  2. 2
    Looks the same. To a person.
  3. 3
    Works for real. Stickers, sounds.

Compare