Data Poisoning

Data poisoning is a type of attack on machine learning models where an adversary intentionally injects misleading or malicious data into the training dataset. This can lead to the model learning incorrect patterns, resulting in poor performance or biased outputs. The primary characteristic of data poisoning is its stealthy nature, as it often goes undetected during the training phase. Common use cases include sabotaging models used in critical applications like finance, healthcare, or security systems. Protecting against data poisoning involves robust data validation techniques and anomaly detection methods to ensure the integrity of training data.