2022 OCT 07 (NewsRx) — By a News Reporter-Staff News Editor at Insurance Daily News — From Alexandria, Virginia, NewsRx journalists report that a patent by the inventors Dakic, Vaso (Irvine, CA, US), Gerritz, Kelly Harold Patrick (Astoria, NY, US), Paquette, Christopher Thomas (New York, NY, US), Perlman, Jennifer Werther (Hillsdale, NJ, US), Romanovski, Pavel (Wallington, NJ, US), Yazovskiy, Anton (Brooklyn, NY, US), filed on December 8, 2021, was printed on-line on September 20, 2022.
The patent’s assignee for patent quantity 11449632 is DeepIntent Inc. (New York, New York, United States).
News editors obtained the next quote from the background info equipped by the inventors: “The approaches described on this part are approaches that could possibly be pursued, however not essentially approaches which have been beforehand conceived or pursued. Therefore, until in any other case indicated, it shouldn’t be assumed that any of the approaches described on this part qualify as prior artwork merely by advantage of their inclusion on this part. Further, it shouldn’t be assumed that any of the approaches described on this part are well-understood, routine, or standard merely by advantage of their inclusion on this part.
“Machine learning techniques have grow to be common for fixing varied forms of issues based mostly on coaching data. A key good thing about a machine learning system is the flexibility to study based mostly on data, bypassing any necessities for handbook coding of an algorithm. Instead, the machine learning system generates an algorithm or mannequin by repeated computations utilizing the coaching data.
“A possible downside of machine learning techniques is that figuring out particular inside working mechanisms of the core machine learning engine may be tough. Most machine learning techniques are configured to generate pretty complicated patterns based mostly on the given coaching data. Because machine learning techniques use complicated algorithms and execute steady learning, figuring out why a machine learning system produced a specific consequence from a set of enter data may be tough, if not unimaginable. In some conditions, this may lead to a lack of accountability; in different conditions, this characteristic protects the coaching data. Because a educated machine learning system exists individually from the coaching data, any data that’s protected or delicate data may be safeguarded throughout using the machine learning system.
“A educated machine learning system inherently protects the data used to train it. However, the coaching section can create points, particularly when the data used to train the machine learning system is strong however protected. Many individuals present data underneath the reassurance that data safety measures will likely be used. As an instance, the Health Insurance Portability and Accountability Act (HIPAA) has stringent necessities on the safety of medical claims data which might forestall a particular person from viewing any of the medical claims data to train a machine learning system.
“Additionally, even when info is protected from viewing, the coaching data or machine learning system can nonetheless present protected info to a viewer. For occasion, a machine learning system utilizing ten inputs might memorize a overwhelming majority of individuals within the United States, thereby offering one-to-one recognition of people as an alternative of offering an algorithm that produces a probability based mostly on common patterns. But to validate the coaching data or the machine learning system would usually contain accessing the coaching data or machine learning system, thereby failing to present the initially desired protections.
“Thus, there’s a want for a system that may protect private, personal, confidential, or in any other case protected info throughout coaching and validation of a machine learning system that makes use of the protected info.”
As a complement to the background info on this patent, NewsRx correspondents additionally obtained the inventors’ abstract info for this patent: “The appended claims might function a abstract of the disclosure.”
The claims equipped by the inventors are:
“1. A pc-implemented methodology comprising: storing, utilizing a server pc executing inside a protected environment, a plurality of media gadgets, every of the media gadgets corresponding to one in every of a plurality of various standing values; receiving, from a requesting computing gadget that’s outdoors the protected environment, a request to ship sure media gadgets outdoors the protected environment to a consumer computing gadget; computing, utilizing a plurality of machine learning techniques executed by the server pc, every of the machine learning techniques having been educated with one of many plurality of standing values as an output, a plurality of probability values related to a specific standing worth for the consumer computing gadget, every of the machine learning techniques having been educated no less than partly by receiving, by the server pc executing inside the protected environment, directions to generate and train a specific machine learning system, utilizing attribute values related to private data information as inputs, and an existence or a non-existence of a one of many plurality of various standing values as outputs, the server pc storing first data comprising a plurality of attribute values for a plurality of the private data information and second data indicating, for every private data file of the plurality of private data information, whether or not the private data file has the standing worth, the server pc being configured to train the actual machine learning system within the protected environment provided that the primary data and the second data fulfill a first criterion and being configured to ship the actual machine learning system to the requesting computing gadget provided that the actual machine learning system satisfies a second criterion; figuring out a specific standing worth, among the many plurality of standing values, having a highest probability worth; choosing a particular set of media gadgets no less than partly based mostly on the recognized specific standing worth having the best probability worth, in a quantity indicated by the request to ship sure media gadgets outdoors the protected environment to the consumer computing gadget; and sending, from the server pc to the consumer computing gadget, the precise set of media gadgets which have been chosen.
“2. The methodology of declare 1, additional comprising the server pc utilizing the best probability worth related to the actual standing worth to dynamically value sending media gadgets to the consumer computing gadget by figuring out a charged value by discounting a customary value by an quantity corresponding to a proportion worth.
“3. The methodology of declare 1, additional comprising the server pc requesting attribute data from an outdoor attribute database based mostly on info acquired from the consumer computing gadget.
“4. The methodology of declare 1, additional comprising: receiving, from the requesting computing gadget that’s outdoors the protected environment, specific attributes for the consumer computing gadget; and figuring out, based mostly on the actual attributes, whether or not to serve a specific media merchandise to the consumer computing gadget.
“5. The methodology of declare 1, additional comprising the server pc storing attribute values for a plurality of various consumer computing gadgets in an attribute database within the protected environment.
“6. The methodology of declare 1, the primary criterion being a minimal variety of cases within the second data of a specific private data file having the standing worth.
“7. The methodology of declare 1, the second criterion being a most fraction of inhabitants in danger.
“8. The methodology of declare 7, additional comprising computing the utmost fraction of inhabitants in danger as a quotient of a variety of cases within the subset of the primary data of a affected person having the standing worth and a variety of constructive predictions of the standing worth from making use of the actual machine learning system to every of the plurality of private data information within the first data.
“9. The methodology of declare 1, additional comprising: coaching the actual machine learning system with a first set of parameters; and figuring out that the actual machine learning system doesn’t fulfill the second criterion and, in response, coaching the actual machine learning system utilizing a second set of parameters.
“10. The methodology of declare 1, the standing being one in every of a specific medical prognosis or a specific prescription.
“11. A pc system comprising: a number of processors; and a number of computer-readable non-transitory storage media coupled to a number of of the processors and storing directions operable when executed by a number of of the processors to trigger the system to carry out a methodology comprising: storing, utilizing a server pc executing inside a protected environment, a plurality of media gadgets, every of the media gadgets corresponding to one in every of a plurality of various standing values; receiving, from a requesting computing gadget that’s outdoors the protected environment, a request to ship sure media gadgets outdoors the protected environment to a consumer computing gadget; computing, utilizing a plurality of machine learning techniques executed by the server pc, every of the machine learning techniques having been educated with one of many plurality of standing values as an output, a plurality of probability values related to a specific standing worth for the consumer computing gadget, every of the machine learning techniques having been educated no less than partly by receiving, by the server pc executing inside the protected environment, directions to generate and train a specific machine learning system, utilizing attribute values related to private data information as inputs, and an existence or a non-existence of a one of many plurality of various standing values as outputs, the server pc storing first data comprising a plurality of attribute values for a plurality of the private data information and second data indicating, for every private data file of the plurality of private data information, whether or not the private data file has the standing worth, the server pc being configured to train the actual machine learning system within the protected environment provided that the primary data and the second data fulfill a first criterion and being configured to ship the actual machine learning system to the requesting computing gadget provided that the actual machine learning system satisfies a second criterion; figuring out a specific standing worth, among the many plurality of standing values, having a highest probability worth; choosing a particular set of media gadgets no less than partly based mostly on the recognized specific standing worth having the best probability worth, in a quantity indicated by the request to ship sure media gadgets outdoors the protected environment to the consumer computing gadget; and sending, from the server pc to the consumer computing gadget, the precise set of media gadgets which have been chosen.
“12. The system of declare 11, the storage media additional comprising directions which when executed by a number of of the processors trigger the system to carry out utilizing the best probability worth related to the actual standing worth to dynamically value sending media gadgets to the consumer computing gadget by figuring out a charged value by discounting a customary value by an quantity corresponding to a proportion worth.
“13. The system of declare 11, the storage media additional comprising directions which when executed by a number of of the processors trigger the system to carry out requesting attribute data from an outdoor attribute database based mostly on info acquired from the consumer computing gadget.
“14. The system of declare 11, the storage media additional comprising directions which when executed by a number of of the processors trigger the system to carry out: receiving, from the requesting computing gadget that’s outdoors the protected environment, specific attributes for the consumer computing gadget; and figuring out, based mostly on the actual attributes, whether or not to serve a specific media merchandise to the consumer computing gadget.
“15. The system of declare 11, the storage media additional comprising directions which when executed by a number of of the processors trigger the system to carry out storing attribute values for a plurality of various consumer computing gadgets in an attribute database within the protected environment.
“16. The system of declare 11, the primary criterion being a minimal variety of cases within the second data of a specific private data file having the standing worth.
“17. The system of declare 11, the second criterion being a most fraction of inhabitants in danger.
“18. The system of declare 17, the storage media additional comprising directions which when executed by a number of of the processors trigger the system to carry out computing the utmost fraction of inhabitants in danger as a quotient of a variety of cases within the subset of the primary data of a affected person having the standing worth and a variety of constructive predictions of the standing worth from making use of the actual machine learning system to every of the plurality of private data information within the first data.
“19. The system of declare 11, the storage media additional comprising directions which when executed by a number of of the processors trigger the system to carry out: coaching the actual machine learning system with a first set of parameters; and figuring out that the actual machine learning system doesn’t fulfill the second criterion and, in response, coaching the actual machine learning system utilizing a second set of parameters.
“20. The system of declare 11, the standing being one in every of a specific medical prognosis or a specific prescription.”
For extra info on this patent, see: Dakic, Vaso. Utilizing a protected server environment to protect data used to train a machine learning system. U.S. Patent Number 11449632, filed December 8, 2021, and printed on-line on September 20, 2022. Patent URL: http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PALL&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.htm&r=1&f=G&l=50&s1=11449632.PN.&OS=PN/11449632RS=PN/11449632
(Our reviews ship fact-based information of analysis and discoveries from world wide.)
https://insurancenewsnet.com/oarticle/patent-issued-for-utilizing-a-protected-server-environment-to-protect-data-used-to-train-a-machine-learning-system-uspto-11449632-deepintent-inc
