New framework halves security regression in Android malware detection

Researchers have identified and quantified a critical issue in continual learning for Android malware detection, termed "security regression." This phenomenon occurs when malware samples that were previously detected by a model evade detection after an update, even if the model's average performance has improved. Experiments revealed that up to 3-6% of malware samples can experience security regression. To address this, a new regression-aware framework, instantiated via Positive Congruent Training (PCT), has been developed. This method effectively halves security regression across various continual learning scenarios while maintaining strong detection performance over time. AI

IMPACT Addresses a critical security vulnerability in continuously updated AI models, potentially improving user trust and detection efficacy.

RANK_REASON Academic paper introducing a new methodology for continual learning in malware detection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Android
• AZ-Class
• Continual Learning
• Daniele Ghiani
• ELSA
• Machine Learning
• Positive Congruent Training
• Tesseract

• paper
• safety
• other

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