Explore a 14-minute conference talk from USENIX Security '23 that addresses the challenge of concept drift in Android malware detection. Learn about innovative methods combining contrastive learning with active learning to combat the rapid decline in effectiveness of machine learning classifiers. Discover how researchers from UC Berkeley developed a new hierarchical contrastive learning scheme and sample selection technique to continuously train Android malware classifiers. Examine the significant improvements achieved by this approach, including reduced false negative and false positive rates, and its ability to maintain consistent performance over an extended period. Gain insights into the evolution of malware and benign apps, and understand the importance of continuous learning in maintaining effective security measures for Android devices.