Data Preprocessing
Ethics & Safety
A type of attack where an adversary manipulates training data to compromise an AI model's performance or behavior.
Detailed Explanation
Data poisoning is an adversarial attack in which malicious actors intentionally alter or inject false data into the training dataset. This manipulation aims to corrupt the learning process, causing the AI model to behave unpredictably, produce incorrect outputs, or make biased decisions. Ensuring data integrity and implementing robust defenses are critical to safeguarding AI systems from such threats.
Use Cases
- •To prevent biased or incorrect results, secure training data by detecting and mitigating data poisoning attacks during model development.
Related Terms
Other terms in the Ethics & Safety category