Data poisoning refers to a method of attack in which an adversary deliberately introduces misleading or harmful data into a maschinellem Lernen (ML) training dataset. The objective of this malicious act is to compromise the integrity of the ML model, leading to incorrect predictions or classifications when the model is deployed in real-world applications.
In vielen maschinellen Lernverfahren systems, the quality and reliability of the Trainingsdaten are crucial for the model’s performance. When an attacker successfully implements data poisoning, they can manipulate the learning process by injecting biased or false information. This can result in a model that performs poorly, behaves unpredictably, or even serves the attacker’s goals by making specific predictions that benefit them.
Datenvergiftung kann verschiedene Formen annehmen, darunter:
- Label-Flipping: Das Ändern der Labels bestimmter Datenpunkte, um das Modell während des Trainings zu täuschen.
- Hintertür-Angriffe: Inserting specific data patterns that cause the model to behave incorrectly only when those patterns are present.
- Ausreißer hinzufügen: Introducing extreme or unusual data points that skew the model’s understanding of the normal Datenverteilung.
Die Minderung von Data Poisoning umfasst mehrere Strategien, wie robuste Datenvalidierung techniques, Anomalieerkennung, and using diverse training datasets to minimize the impact of any single source of data corruption. Continuous monitoring and updating of models can also help to reduce the risk of data poisoning attacks.