PhD Research Seminar by LIU Huiwen | Federated learning: a survey on challenges, methods and future directions

Federated learning has been a hot research topic in enabling the collaborative training of machine learning models among different organizations under the privacy restrictions. As researchers try to support more machine learning models with different privacy-preserving approaches, there is a requirement in developing systems and infrastructures to ease the development of various federated learning algorithms. Meanwhile, there is a central server to coordinate the training process in the traditional centralized federated learning model, which has a fatal drawback that If there is a problem with the central server, the whole system will be affected. So, we proposed a new problem named Decentralized Byzantine Federated Learning (i.e., DBFL) and design a novel federated Decentralized Byzantine fault-tolerant learning framework named FedDBFT to solve the problem. In this framework, we design a novel consensus mechanism named Proof of Data Contribution (i.e., PoDC) to solve the distributed consensus problem to make the whole system to train a common model. Meanwhile, we design an incentive mechanism to motivate all participating nodes cooperate to train a model with high generalization and robust performance in the case of no central organization node and Byzantine nodes.