[Privacy Computing Written Talk] The history and application of secure multi-party computing

In 1982, Academician Yao Qizhi put forward the problem of calculating the security of both parties through the "Millionaire Problem". The "Millionaire Problem" refers to how two millionaires can compare who is richer between the two without the participation of a third party:

The safe calculation of the two parties can be widely explained as: there are two people Alice and Bob, Alice masters the number a, and Bob masters the number b. How to use the numbers a and b together without telling each other about the specific values of the numbers a and b b Perform calculations.

While raising the "Millionaire Problem", Academician Yao Qizhi gave three solutions and discussed the application of Secret Vote, Oblivious Negotiation, and Private Querying of Database. .

Afterwards, Goldreich discussed Secure Multi-Party Computation in 1987, and proposed a secure multi-party computing protocol that can calculate arbitrary functions in the sense of computing. Goldreich also proved theoretically that all secure multi-party computing protocols can be realized through Universal Circuit valuation. Then in 1988, Goldreich summarized and defined the security of multi-party computing.

Later in 1989, Beaver and others studied the problem of secure multi-party scientific computing under the information-theoretic security model, and proposed a secure multi-party arithmetic operation protocol that can achieve information-theoretic security with a constant round of complexity.

Secure multi-party computing has both theoretical research and practical application value. has broad application prospects in electronic voting, privacy protection data mining, machine learning, blockchain, biological data comparison, cloud computing and other fields.

Voting elections in real life ensure fairness and justice through the uniform use of blank ballots, ballot boxes, credible tellers, and live video recording throughout the entire process. In the field of electronic voting, when voters vote at home, the computer at home may have been infected with viruses, and the voting results may be maliciously obtained and tampered with. Therefore, the electronic voting system must ensure that voters know whether their voting information has been submitted correctly and whether it has been submitted. Malicious attackers tamper with, and at the same time, protect the voter's voting information from being obtained by anyone other than the voter. Secure multi-party computing provides a good solution to the problem of how to protect private information and ensure the correctness of results in this distributed environment.

Cramer et al. proposed the first multiple-choice electronic voting scheme based on ElGamal threshold encryption technology and zero-knowledge proof. Later, Damgard et al. proposed a multiple-choice electronic voting scheme based on Pailier homomorphic encryption technology. In 1992, A. Fujioka et al. proposed the famous FOO electronic voting protocol using blind signature technology.

As a very effective data analysis tool, data mining can discover hidden laws in data, and has important applications in scientific and policy research, business decision-making and other aspects. However, the data to be mined often contains a lot of sensitive information, so it must be protected and data mining is carried out under privacy protection. When conducting data mining in a multi-party situation, participants are often unwilling to share data, but only willing to share the results of data mining. This situation is very common in scientific and medical research. For example, the patient information of various medical institutions is sensitive information. Will be willing to disclose. Application security multi-party computing can complete data mining through multi-party collaboration while protecting the data and information of all parties from being leaked.

Machine learning has been applied to various fields, causing a lot of changes, such as image and speech recognition, anomaly detection, etc. In order to achieve good results in machine learning, a large amount of data is needed for model training. The privacy protection of training data is also a problem. When multiple institutions cooperate for model training, the data is distributed among different participants, and secure multi-party computing can allow each institution to successfully conduct model training while protecting the privacy of sensitive data.

In short, when each participant is in a distributed environment and there are requirements for data privacy protection, it is very suitable for multi-party application security.