Adversarial example
An input crafted with small, often imperceptible perturbations that causes a model to misclassify or misbehave, exploiting the brittleness of learned decision b
An input crafted with small, often imperceptible perturbations that causes a model to misclassify or misbehave, exploiting the brittleness of learned decision boundaries.
An input crafted with small, often imperceptible perturbations that causes a model to misclassify or misbehave, exploiting the brittleness of learned decision boundaries.
evolution
- 2004 · historyDiscovery of adversarial spam filtering
Dalvi et al. demonstrated that machine learning-based spam filters could be bypassed by attackers modifying email content.
- 2013 · historyFormalization of adversarial examples
Szegedy et al. identified that deep neural networks are highly susceptible to small, non-random perturbations that cause misclassification.
- 2014 · historyFast Gradient Sign Method (FGSM)
Goodfellow et al. introduced a computationally efficient method to generate adversarial examples, highlighting the linear nature of neural networks.
- 2016 · historyUniversal adversarial perturbations
Moosavi-Dezfooli et al. proved that a single, image-agnostic perturbation can cause misclassification across a wide range of inputs.
- 2017 · historyPhysical world attacks
Researchers demonstrated that adversarial examples could be printed or applied to physical objects to fool autonomous systems in real-world conditions.