Research on Efficient Deep Learning Method Based on Blockchain Consensus Mechanism

Aiming at the problem of trustworthy model aggregation for distributed deep learning in blockchain platforms, this paper proposes a deep learning algorithm based on the efficient consensus mechanism of blockchain. The algorithm improves the efficiency of model aggregation by improving the Byzantine fault-tolerant consensus algorithm based on the voting mechanism and uses the Krum algorithm based on differential privacy to improve the convergence of model aggregation. Through experiments, it is shown that the communication complexity is reduced from
$\mathrm{O}(2\mathrm{n}^{2})$
to
$\mathrm{O}(\mathrm{n}^{2})$
, and the aggregation model is still able to converge and maintain a high classification accuracy with 30% malicious nodes