BrainScape: Research on Sovereign Cloud Architecture of Brain-Computer Interface Based on Distributed Weight Merging and DARE Algorithm

Brain-computer interfaces (BCI) face two core bottlenecks: neural data privacy leakage and poor model generalization due to EEG data heterogeneity (Non-IID). Traditional schemes struggle to balance privacy, performance, and heterogeneous adaptation, lacking "neural sovereignty" protection. This paper proposes the BrainScape distributed sovereign cloud architecture, integrating federated learning, zk-SNARKs, BIAS correction, TIES-Merging, and DARE algorithm to realize "data available but invisible". Experimental results show its cross-user accuracy reaches 89.3%, member inference attack success rate drops to 52%, and it tolerates one-third of malicious nodes, providing technical and engineering support for BCI industrialization.