Which term best describes a system's susceptibility to adversarial examples that humans would recognize as incorrect?

Get ready for the GARP Risk and AI Exam with flashcards and multiple choice questions. Each question comes with hints and explanations. Prepare for success!

Multiple Choice

Which term best describes a system's susceptibility to adversarial examples that humans would recognize as incorrect?

Vulnerability is the idea that a system has a weakness that can be exploited by crafted inputs, revealing why it can fail in surprising ways. In machine learning, adversarial examples are tiny, often imperceptible perturbations designed to push the model to make a wrong prediction. The striking point is that humans would recognize the input as incorrect, yet the model still errs, showing a gap between human perception and the model's learned patterns. The other terms describe how models are trained or learned (back-propagation as the training algorithm, supervised or unsupervised learning as learning paradigms) and don’t capture the system’s susceptibility to such deceptive inputs.

Vulnerability is the idea that a system has a weakness that can be exploited by crafted inputs, revealing why it can fail in surprising ways. In machine learning, adversarial examples are tiny, often imperceptible perturbations designed to push the model to make a wrong prediction. The striking point is that humans would recognize the input as incorrect, yet the model still errs, showing a gap between human perception and the model's learned patterns. The other terms describe how models are trained or learned (back-propagation as the training algorithm, supervised or unsupervised learning as learning paradigms) and don’t capture the system’s susceptibility to such deceptive inputs.

More practice questions