Crisis analysis on blockchain-based decentralized learning in wireless networks

The nodes may perform data training and make decisions according to the learning results for better performance in wireless networks. However, to guarantee privacy-preserving service, learning model parameters are exchanged among the nodes instead of exchanging the original raw data obtained at each node. Although the information exchange could be performed in a distributed manner by exploiting blockchain for verification, there still exist problems of jamming attacks in wireless channels. The jammers may interrupt transactions requests toward the normal nodes works as miners or try to corrupt the information. In this paper, we provide analysis results that blocks originated from attackers with jamming capabilities becomes main stream. Through numerical results, we show that attacks to normal miners are highly likely to be succeed by the attackers with jamming abilities even though the computing power sum of attackers is less than 51% in wireless networks.