Learn in the textbookDenial-of-Service, Zero-Day, and Attacks Targeting AI
DefinitionAn adversarial example (敵対的サンプル) is input data altered in ways people can barely detect, crafted to make an AI machine learning model judge wrongly.
ExplanationA machine learning model is a system that learns rules from large amounts of data and then classifies or judges new input. For example, if an image recognition model is given a picture with a tiny amount of noise added, it may identify it as something else even though it looks unchanged to people.
Memory hook"Adversarial" signals that the data was made to fool the AI. "Example" here means a single piece of data fed to the model.