Currently cyberinfrastructures are facing increasinglystealthy attacks that implant malicious payloads under thecover of benign programs. Existing attack detection approachesbased on statistical learning methods may generate misleadingdecision boundaries when processing noisy data with such amixture of benign and malicious behaviors. On the other hand,attack detection based on formal program analysis may lackcompleteness or adaptivity when modeling attack behaviors.In light of these limitations, we have developed LEAPS, anattack detection system based on supervised statistical learningto classify benign and malicious system events. Furthermore,we leverage control flow graphs inferred from the system eventlogs to enable automatic pruning of the training data, whichleads to a more accurate classification model when applied tothe testing data. Our extensive evaluation shows that, comparedwith pure statistical learning models, LEAPS achieves consistentlyhigher accuracy when detecting real-world camouflaged attackswith benign program cover-up. About the speaker: Kexin Pei is a second year master student at Department of Computer Science, Purdue University. His research interests include data mining and security, focusing on solving security problems using program analysis and machine learning techniques.