Private analytics using multi-party computation
View Patent ↗A method, system, and computer program product for performing strong desensitization of sensitive data within a garbled circuit includes: compiling a predetermined program into a first program, where the compiled first program is encoded in a form of a garbled circuit, and where the predetermined program runs on sensitive data; and executing the first program, where executing the first program includes: executing an analytics function using tokenized data with a first set of sensitive information and analytics data with a second set of sensitive information, where the tokenized data originated from a data provider and the analytics data originated from an analytics provider; and generating an output of the first program using a result of the analytics function, where the output contains desensitized data.
1. A computer-implemented method comprising:
compiling a predetermined program into a first program, wherein the compiled first program is encoded in a form of a garbled circuit, and wherein the predetermined program runs on sensitive data; and
executing the first program, wherein a first matched half of the first program is executed on the analytics provider and a second matched half of the first program is executed on the data provider, and wherein executing the first program comprises:
executing an analytics function using tokenized data with a first set of sensitive information and analytics data with a second set of sensitive information, wherein the tokenized data originated from a data provider and the analytics data originated from an analytics provider, and
generating an output of the first program using a result of the analytics function, wherein the output contains desensitized data.
2. The method of claim 1 , wherein executing the first program further comprises:
in response to executing the analytics function, desensitizing the result of the analytics function using a statistical desensitization technique, wherein the desensitizing comprises obscuring the first set of sensitive information and the second set of sensitive information.
3. The method of claim 1 , wherein executing the first program further comprises encrypting the output.
4. The method of claim 3 , wherein:
one of the data provider and the analytics provider is an owning party and one is a non-owning party, and
encrypting the output comprises:
receiving an encryption key from the owning party, and
encrypting the output using the encryption key.
5. The method of claim 4 , wherein the owning party uses the encryption key to decrypt the encrypted output.
6. The method of claim 1 , wherein only the analytics provider receives the output.
7. The method of claim 1 , wherein the analytics provider and the data provider receive the output.
8. The method of claim 1 , wherein:
the data provider provides the tokenized data by encoding the tokenized data into the second half of the first program, and
the analytics provider provides the analytics data by encoding the analytics data into the first half of the first program.
9. The method of claim 1 , wherein the tokenized data includes sensitive data elements that are replaced with tokens, such that the sensitive data elements are obscured.
10. The method of claim 1 , The method of claim 1 , wherein:
the analytics function is a neural net, and
the second set of sensitive information includes at least weights and biases of the neural net.
11. The method of claim 1 , wherein at least one of the first matched half of the first program and the second matched half of the first program run on a cloud infrastructure.