Secure and Efficient Gossip Learning with Adversarial Nodes

Gossip learning is one of the most advanced approaches in decentralized techniques and performs almost as well as centralized training. However, gossip learning is inherently vulnerable to malicious attacks from adversarial nodes due to its flat architecture, which leads to performance degradation of collaborative training nodes. To reduce the negative impact of adversarial nodes, we propose an effective strategy to protect the distributed training process. The results show that the proposed strategy is robust to malicious attacks.