Automatic privacy-aware machine learning method and apparatus

Patent metadata

Field	Value
Publication number	US-11438314-B2
Application number	US-202117172679-A
Country	US
Kind code	B2
Filing date	Feb 10, 2021
Priority date	Feb 10, 2021
Publication date	Sep 6, 2022
Grant date	Sep 6, 2022

Abstract

Disclosed are systems and methods for improving interactions with and between computers in content generating, searching, hosting and/or providing systems supported by or configured with personal computing devices, servers and/or platforms. The systems interact to identify and retrieve data within or across platforms, which can be used to improve the quality of data used in processing interactions between or among processors in such systems. The disclosed systems and methods provide systems and methods for privacy-aware machine learning using an improved data encoding that withholds more information. The disclosed systems and methods encode a corpus of data and encode each query used in searching and generating query results from the corpus of encoded data.

First claim

The invention claimed is:

1. A method comprising:
storing, using a data store accessible by a computing device, sensitive data, the sensitive data comprising a plurality of records, each record, of the plurality of records, comprising a plurality of unencoded values corresponding to a plurality of fields of the record;
determining, via a computing device and using the sensitive data, a set of unencoded values for each field of the plurality of fields;
determining, via the computing device, a set of encoded values exclusive to each field of the plurality of fields, the determining comprising, for each field of the plurality using the set of unencoded values determined for the field;
generating, via the computing device and for each field of the plurality of fields, a mapping between the set of encoded values exclusive to the field and the set of unencoded values determined for the field, each unencoded value in the set of encoded values having a corresponding encoded value;
generating, via the computing device, a plurality of encoded records, the generating comprising replacing each unencoded value in the plurality of records with an encoded value based on a field in the plurality of fields corresponding to the unencoded value and the mapping determined for the corresponding field;
transmitting, via the computing device and to a service provider, the plurality of encoded records;
generating, via the computing device, an encoded query using an unencoded query, each unencoded value of the unencoded query corresponding to a field of the plurality of fields, the generating comprising replacing each unencoded value from the unencoded query with an encoded value based on the field of the plurality of fields corresponding to the encoded value and the mapping generated for the corresponding field;
transmitting, via the computing device, the encoded query to the service provider; and
receiving, via the computing device and from the service provider, a response to the encoded query.

2. The method of claim 1 , generating the plurality of encoded records further comprising:
randomly shuffling an ordering of the plurality of fields in each encoded record of the plurality of encoded records.

3. The method of claim 1 , the encoded query comprising a set of encoded input field values corresponding to a set of input fields of the encoded query and at least one encoded label field value corresponding to at least one label field of the encoded query, each input field in the set of input fields and the at least one label field corresponding to one field of the plurality of fields.

4. The method of claim 3 , generating the encoded query further comprising:
randomly shuffling an ordering of the set of input fields of the encoded query; and
randomly shuffling an ordering of the at least one label field of the encoded query.

5. The method of claim 3 , the response comprising an estimate corresponding to the at least one label field of the encoded query, the estimate comprising an indication whether the encoded value of the at least one label field can be inferred given the encoded values of the set of input fields of the encoded query.

6. The method of claim 5 , the estimate being determined based on a frequency of occurrence of the encoded value of the at least one label field and a number of encoded records, from the plurality of encoded records, used to determine the frequency.

7. The method of claim 6 , the frequency and the number of encoded records used to determine the frequency are received in an encrypted form.

8. The method of claim 7 , the number of encoded records used to determine the frequency is at least equal to a minimum threshold.

9. The method of claim 3 , further comprising:
determining, for each encoded record of the plurality of encoded records, distance metric value indicating a number of the set of encoded input field values of the encoded query missing from a set of encoded field values of the encoded record; and
determining, for each encoded record of the plurality of encoded records, a frequency measure indicating whether or not the at least one label field value exists in the set of encoded field values of the encoded record.

10. The method of claim 9 , further comprising:
selecting a set of encoded records from the plurality of encoded records, each encoded record from the selected set has a distance metric value from a set of distance metric values selected in accordance with a minimum threshold number of encoded records to be included in the set of encoded records; and
determining the frequency using the frequency measure determined for each encoded record included in the selected set of encoded records.

11. The method of claim 10 , the set of distance values is iteratively determined with a first distance metric value selected representing a least distance metric value and any distance value metric selected thereafter represents a lesser distance metric value than any unselected distance metric value.

12. A non-transitory computer-readable storage medium tangibly encoded with computer-executable instructions that when executed by a processor associated with a computing device perform a method comprising:
storing, using a data store, sensitive data, the sensitive data comprising a plurality of records, each record, of the plurality of records, comprising a plurality of unencoded values corresponding to a plurality of fields of the record;
determining, using the sensitive data, a set of unencoded values for each field of the plurality of fields;
determining a set of encoded values exclusive to each field of the plurality of fields, the determining comprising, for each field of the plurality using the set of unencoded values determined for the field;
generating, for each field of the plurality of fields, a mapping between the set of encoded values exclusive to the field and the set of unencoded values determined for the field, each unencoded value in the set having a corresponding encoded value;
generating a plurality of encoded records, the generating comprising replacing each unencoded value in the plurality of records with an encoded value based on a field in the plurality of fields corresponding to the unencoded value and the mapping determined for the corresponding field;
transmitting, to a service provider, the plurality of encoded records;
generating an encoded query using an unencoded query, each unencoded value of the unencoded query corresponding to a field of the plurality of fields, the generating comprising replacing each unencoded value from the unencoded query with an encoded value based on the field of the plurality of fields corresponding to the encoded value and the mapping generated for the corresponding field;
transmitting the encoded query to the service provider; and
receiving, from the service provider, a response to the encoded query.

13. The non-transitory computer-readable storage medium of claim 12 , generating the plurality of encoded records further comprising:
randomly shuffling an ordering of the plurality of fields in each encoded record of the plurality of encoded records.

14. The non-transitory computer-readable storage medium of claim 12 , the encoded query comprising a set of encoded input field values corresponding to a set of input fields of the encoded query and at least one encoded label field value corresponding to at least one label field of the encoded query, each input field in the set of input fields and the at least one label field corresponding to one field of the plurality of fields.

15. The non-transitory computer-readable storage medium

Assignees

• Yahoo Assets Llc

Inventors

• Bax Eric
• Bax Charlotte

Classifications

• H04L63/0428 Primary

  wherein the data content is protected, e.g. by encrypting or encapsulating the payload · CPC title

• G06N20/00

  Machine learning · CPC title