An auditable privacy protection deep learning platform construction method based on a block chain incentive mechanism

Abstract

The invention discloses an auditable privacy protection deep learning platform construction method based on a block chain incentive mechanism. The problem of parameter deficiency in deep learning model training is solved, so that trainers of a plurality of similar models can cooperate to train the deep learning model under the condition that privacy is protected and correctness of the shared parameters can be audited. The technical effects obtained by the method are as follows: firstly, the privacy of parameters is ensured by an encryption method used by a model trainer, and all participants are required to cooperate in the process of decrypting the updated parameters, so that the possibility of parameter leakage is further reduced; Secondly, the encrypted parameters are stored in the block chain in a state form, and only participants and authorized miners can access the encrypted parameters; Thirdly, due to the existence of an excitation mechanism based on a block chain, the effectiveness of parameters is ensured; And the participant needs to pay the collateral money when submitting the parameters, and if the parameters are invalid, the collateral money is not received, so that the auditability of the shared parameters is ensured.

Description

technical field [0001] The invention relates to the technical field of cyberspace security, in particular to an auditable privacy protection deep learning platform construction method based on a block chain incentive mechanism. Background technique [0002] (1) Deep Learning [0003] Deep learning is a branch of machine learning. It is a technology to realize machine learning. The most basic method of machine learning is to use algorithms to analyze data, learn from it, and then make decisions and predictions about events in the real world. Learning The best method is to use a large amount of data to "train" and learn how to complete tasks from the data through various algorithms. [0004] The training of deep learning is actually training the neural network. A neural network is actually a plurality of neurons connected according to certain rules, and neurons are arranged according to layers. The leftmost layer is called the input layer, which is responsible for receiving...

AI Technical Summary

Problems solved by technology

[0008] Consistency problem: The Bitcoin blockchain solves the consistency problem of the decentralized bookkeeping system through competitive bookkeeping

[0012] For ordinary encryption algorithms, the data can be encrypted into ciphertext to protect the privacy of the data, but if two encrypted data are calculated, and then the calculation result is decrypted, the result is generally meaningless code

Images