Adversarial Robustness

Quick Definition

AI model ability to resist adversarial perturbations

The ability of AI models to maintain correct predictions when facing adversarial perturbations or attacks.

adversarial training, certified defenses, robust optimization

adversarial-attack ai-safety

Running AI on local devices instead of the cloud

Reducing model precision from 32-bit to lower bit representations

AI that enables machines to interpret visual information

AI using symbolic logic and explicit knowledge rules

Quantifying each input features contribution to predictions

Converting written text into natural-sounding spoken audio