Visa International Service Association

Ara Jermakyan

Ajit Gaddam

Pushkar Joglekar

Vlad Spichek

Methods and systems for inducing model shift in a malicious computer's machine learning model is disclosed. A data processor can determine that a malicious computer uses a machine learning model with a boundary function to determine outcomes. The data processor can then generate transition data intended to shift the boundary function and then provide the transition data to the malicious computer. The data processor can repeat generating and providing the transition data, thereby causing the boundary function to shift over time.

Proactive defense of untrustworthy machine learning system

Methods and systems for analyzing request to access resources and determining a resource access policy are provided. The resource access system can train, store, evaluate, and deploy machine learning models that learn to output a trust score associated with a resource access request, the trust score relating to the request's legitimacy. A resource access system can receive a request for a resource from a requesting entity. The resource access system can determine an entity profile using request data in the request. The resource access system can request data from the request to determine whether the request is normal or anomalous. Using a policy engine, the resource access system can determine a resource access policy, such as allowing or denying access to the resource.

Real-time entity anomaly detection

Disclosed are a method and system for generating a behavior profile for a new entity. The method is performed in response to and immediately after a new entity is added to a fraud detection system. The method includes generating a doppelganger behavior profile for the new entity based on feature data of the new entity and pre-existing entities. Generating the profile includes determining a similarity score for each pre-existing entity, ranking the plurality of pre-existing entities based on the score, selecting pre-existing entities based on the ranking, and combining behavior profiles of the preexisting entities to produce the profile. The method also includes assigning the profile to the new entity and monitoring the new entity. The method further includes detecting fraud, an anomaly, and/or malicious intent based on the profile, and automatically suspending an aspect of the new entity in response to the detection.

Systems and methods for generating behavior profiles for new entities

Methods and systems for evaluating microservice system level activities including system calls and commands, and generating security policies for microservices are disclosed. A microservice agent, operating on a microservice host, can collect system level activity data corresponding to a plurality of microservices operating on the microservice host. The microservice agent can transmit the system level activity data to a microservice evaluator that can use the system level activity data to train machine learning models to identify normal and abnormal microservice system level activities. The normal and abnormal system level activities can be used to generate security policies that can be applied to the microservices. Microservices that perform abnormal system level activities or system level activities that violate security policies can be paused or terminated.

Microservice adaptive security hardening

A method and system for protecting access to remote systems, such as resource databases containing sensitive resources, such as cryptographic keys or personally identifying information, is disclosed. A server can receive a plurality of access requests from a requesting entity. The server can identify an entity profile corresponding to the access requests, as well as a machine learning model corresponding to the entity profile. The access requests can be used to form access sequences, which can be evaluated by the machine learning model. The machine learning model returns an anomaly score that can be compared to a threshold. If the anomaly score exceeds the threshold, the server can prevent further access to the remote system, for example, by revoking a credential associated with the requesting entity.

Systems and methods to secure API platforms

A method and system for generating a behavior profile for a new entity to be added to a behavior-monitored system include receiving feature data of a new entity and feature data of a plurality of pre-existing entities from at least one data source. The plurality of pre-existing entities are associated with a plurality of pre-existing behavior profiles. A doppelganger behavior profile is determined for the new entity based at least partially on the feature data of the new entity and the feature data of the plurality of pre-existing entities, and the doppelganger behavior profile is assigned to the new entity. A target action is implemented with respect to the new entity based at least partially on the doppelganger behavior profile assigned to the new entity.

Growing community of inventors

Northridge, CA, United States of America

Ara Jermakyan