Mastering Machine Learning for Penetration Testing

Poisoning attacks

In machine learning poisoning attacks, attackers poison the model in order to change the learning outcome, by adding malicious data in the model training phase. This method can be performed, for example, by sending and injecting carefully designed samples when data collection is occurring during network operations, to train a network intrusion detection system model. The following workflow illustrates how a poisoning attack occurs:

Some of the greatest research conducted on adversarial machine learning was done in the Pattern recognition and applications lab Italy, including Poisoning Attacks Against Support Vector Machines, ...