Collaborative multi-parties/multi-sources machine learning for affinity assessment, performance scoring, and recommendation making

Abstract

Provided is a process that includes sharing information among two or more parties or systems for modeling and decision-making purposes, while limiting the exposure of details either too sensitive to share, or whose sharing is controlled by laws, regulations, or business needs.

What is claimed is: 
     
         1 . A tangible, non-transitory, machine-readable medium, storing instructions that when executed by one or more processors effectuate operations comprising:
 obtaining, from a first computer system, at a second computer system, an output of an upstream machine learning model executed by the first computer system and a value that corresponds to a subset of entities having respective records in a data repository accessible to the second computer system;   retrieving, with the second computer system, based on the value, from the data repository, features of members of the subset of entities indicated in corresponding records in the data repository;   inferring, with a downstream machine learning model executed by the second computer system, based on the output of the upstream machine learning model and the retrieved features of members of the subset of entities, a property of the a member of the subset of entities; and   storing, with the second computer system, the property in association with the member of the subset of entities in memory.   
     
     
         2 . The medium of  claim 1 , wherein:
 the second computer system does not have access to at least some input features of the upstream machine learning model upon which the output is based.   
     
     
         3 . The medium of  claim 1 , wherein:
 the subset of entities are a subset of people having profiles in the data repository; and   the subset is less than 10% of the people having profiles in the data repository.   
     
     
         4 . The medium of  claim 1 , wherein:
 the value uniquely identifies an entity among entities having respective records in the data repository; and   at least some information in a record of the uniquely identified entity is not available to the first computer system.   
     
     
         5 . The medium of  claim 1 , wherein:
 the output is a token; and   the value comprises a token-context value.   
     
     
         6 . The medium of  claim 1 , wherein:
 the output comprises a vector output by an autoencoder of the first computer system, the autoencoder being trained to map higher-dimensional inputs in records available to the first computer system but not the second computer system into lower-dimensional vectors that preserve at least some information in records available to the first computer system without revealing all input features to the autoencoder.   
     
     
         7 . The medium of  claim 1 , wherein:
 the upstream machine learning model and the downstream machine learning models are trained separately.   
     
     
         8 . The medium of  claim 1 , wherein:
 the upstream machine learning model and the downstream model are jointly trained.   
     
     
         9 . The medium of  claim 1 , wherein:
 the upstream machine learning model and the downstream machine learning model cooperate to infer affinity of an entity for a product or service without sharing at least some input features of the upstream machine learning model with the second computer system.   
     
     
         10 . The medium of  claim 9 , wherein:
 at least some input features of the downstream machine learning model are not shared with the first computer system.   
     
     
         11 . The medium of  claim 1 , wherein:
 no input features of the upstream machine learning model upon which the output is based are communicated to the second computer system, other than the value to the extent the value is an input feature.   
     
     
         12 . The medium of  claim 1 , wherein:
 the upstream machine learning model and the downstream machine learning model cooperate to score performance of an entity for a product or service without sharing at least some input features of the upstream machine learning model with the second computer system.   
     
     
         13 . The medium of  claim 12 , wherein:
 the score is one of a sequence of scores over time in an iterated risk assessment.   
     
     
         14 . The medium of  claim 13 , wherein:
 the iterated risk assessment is a continuous risk assessment.   
     
     
         15 . The medium of  claim 12 , wherein:
 the performance score is indicative of cybersecurity risk.   
     
     
         16 . The medium of  claim 1 , wherein:
 the upstream machine learning model and the downstream machine learning model cooperate to recommend a product or service for an entity without sharing at least some input features of the upstream machine learning model with the second computer system.   
     
     
         17 . The medium of  claim 1 , wherein:
 the second machine learning model comprises means for machine learning; and   the operations comprise providing, from the second computer system, to a third computer system, the property and another value that corresponds to a subset of entities having respective records in a data repository accessible to the third computer system.   
     
     
         18 . The medium of  claim 1 , wherein:
 the output is token corresponding to a principle component of input features of the first machine learning model.   
     
     
         19 . The medium of  claim 1 , wherein:
 the output is based on a combination of a response of the first machine learning model to a set of input features and noise that obfuscates the set of input features and the response while causing at least some population statistics of a set of outputs of the first machine learning model to change by less than 10% relative to a set of outputs that are not combined with the noise.   
     
     
         20 . A method, comprising:
 obtaining, from a first computer system, at a second computer system, an output of an upstream machine learning model executed by the first computer system and a value that corresponds to a subset of entities having respective records in a data repository accessible to the second computer system;   retrieving, with the second computer system, based on the value, from the data repository, features of members of the subset of entities indicated in corresponding records in the data repository;   inferring, with a downstream machine learning model executed by the second computer system, based on the output of the upstream machine learning model and the retrieved features of members of the subset of entities, a property of the a member of the subset of entities; and   storing, with the second computer system, the property in association with the member of the subset of entities in memory.