Forest Global Training Partners

Cyber Risks · Unit 42 Palo Alto Networks

The Contrastive Credibility Propagation Algorithm in Action: Improving ML-powered Data Loss Prevention

Summary: Researchers at Palo Alto Networks developed Contrastive Credibility Propagation (CCP), a semi-supervised learning algorithm presented at the 38th AAAI Conference on Artificial Intelligence, which the paper shows is robust to five data quality issues found in real-world datasets. The post applies CCP to a machine learning DLP classifier that separates sensitive from non-sensitive text documents and identifies the type of sensitive data, aiming to reduce accuracy loss when moving from a curated test set to real-world data.

Why it matters: Real-world DLP traffic is noisy and unviewable due to privacy concerns, and pseudo-labels generated during semi-supervised training inevitably contain errors that deep neural networks are highly sensitive to. The paper's stated goal is closing the accuracy gap between a curated test set and real-world data, which is where a DLP classifier either holds up or fails.

Read the original at Unit 42 Palo Alto Networks ↗

Topics
AI Security
Related