Adversarial Machine Learning

Adversarial machine learning studies attacks and defenses involving learned models. An adversarial example targets inference, while data poisoning corrupts the training process.

Japanese terms

1. Adversarial machine learning — 敵対的機械学習てきたいてききかいがくしゅう: The study of attacks on machine-learning systems and defenses against them.
2. Adversarial example — 敵対的てきたいてきサンプル: An input deliberately modified to cause a model to make an incorrect prediction.
3. Data poisoning — データポイズニング: An attack that corrupts training data to influence a model’s learned behavior.

Related topics

AI safety and evaluation includes adversarial testing and robustness measurement.