Adversarial meaning: Adversarial machine learning is a machine learning approach that uses adversarial information to trick or mislead a model. While adversarial machine learning can be used for various purposes, it is typically used to carry out an attack or cause a machine learning system to malfunction. An instance can be modified to run on many models with different datasets or architectures.
Adversarial machine learning is classified as a white box or black box assault. In a white box attack, the attacker is aware of the internal workings of the model, while in a black box attack, the attacker is only aware of the output of the model.
Large datasets relevant to the topic being learned are used to train machine learning models. For example, suppose a car manufacturer wanted to teach its self-driving car to recognize a stop sign. In that case, they could use a machine learning algorithm to feed thousands of photos of stop signs. An adversary machine learning attack could be using that machine learning algorithm against you, leveraging the algorithm's input data (in this example, photos of stop signs) to misinterpret it, leading the entire system to misidentify stop signs. high when implemented in practice or production.
