NDSS 2025 – Density Boosts Everything

This piece summarizes a research paper presented at NDSS 2025 that introduces a one-stop strategy to enhance AI-based malware detectors by addressing feature sparsity. Through a novel compression technique and a density-boosting training method, the approach aims to improve performance, robustness against poisoning and evasion attacks, and long-term sustainability amid concept drift. The authors report empirical gains and position the methods as complementary to existing defenses.