Data poisoning attacks evaluated on supervised learning models

A new research paper published on arXiv evaluates the effectiveness of two data poisoning attacks, label flipping and backdoor poisoning, on common supervised learning models. The study found that label flipping significantly degraded the performance of Logistic Regression and Linear SVM models, while Random Forest remained more stable. Backdoor poisoning, however, proved highly effective, achieving near-perfect attack success rates across all tested models and datasets while maintaining high clean-test accuracy, highlighting the stealthy nature of such targeted attacks. AI

IMPACT Highlights the stealth and effectiveness of backdoor poisoning attacks, underscoring the need for advanced security evaluations beyond standard metrics for AI models.

RANK_REASON The cluster contains a research paper detailing empirical evaluations of data poisoning attacks on supervised learning models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Fashion-MNIST
• Linear Svm
• logistic regression model
• MNIST database
• random forest

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →