Securing ML-based Android Malware Detectors: A Defensive Feature Selection Approach against Backdoor Attacks

Project: Securing ML-based Android Malware Detectors: A Defensive Feature Selection Approach against Backdoor Attacks

• The project was implemented in association with NASK (National Research Institute).
• Comparison of different attacks on machine-learning models on classic and federated-learning scenarios with improved attacks in more realistic scenarios.
• Proposal of an effective feature selection method to reduce the vulnerability of a machine learning system to a backdoor attack.