A distributed machine learning dynamic remote proof scheme that resists collusion attacks

Based on the security of distributed machine learning data and model based on remote proof, a cross-domain dynamic remote proof scheme based on ring signature improves the efficiency of computer cluster proof, the application of dynamic signature, realizes the real-time measurement of the confidence of computer cluster, and strengthens the real-time protection of data and model. Computers with different security domain may adopt different remote proof mechanism, and convert the proof information of different proof methods into unified proof information, thus realizes the cross-domain dynamic remote proof between computer clusters. Noise is added when the cloud server distributes data to a distributed computer cluster. Due to the particularity of distributed architecture and the unsolution of indefinite equations, the enemy cannot conspire with one or more computers to steal data or the model parameters trained by endpoints. Finally, the safety of this method is verified through safety analysis and experimental analysis, and the computational overhead required in the proof process is reduced compared with other methods.