Evasion attacks vs Poisoning attacks

Evasion attacks refer to techniques used to bypass or deceive AI systems, particularly in security contexts where malicious actors attempt to avoid detection by altering input data. This concept is often studied in adversarial machine learning to improve model robustness.

Poisoning attacks refer to adversarial techniques where malicious actors intentionally corrupt training data to compromise the performance or behavior of machine learning models. This can involve injecting false data or manipulating existing data to bias the model's outcomes.