Federated Learning Robustness in Android Malware Detection

This project is about improving Federated Learning (FL). FL is a machine learning technique that trains an algorithm on multiple decentralised clients or devices that hold local data samples without exchanging them. In this system, since clients do not have to share their private data during training, their privacy is preserved. In FL, there is a server that aggregates the client’s local models by using an aggregation function. Poisoning attacks is one of the threats of this system, which the attacker/adversary can poison some local models by changing their parameters and send poisoned models to the server. In this way, the final aggregated model will also be poisoned. This project will improve the aggregation function in federated learning systems to be robust against poisoning attacks.