Bootstrapping Access Models for New Users
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Access control systems face challenges in building reliable access models for new users due to insufficient training data, leading to unreliable detection of unusual or anomalous behavior.
Innovation Solution
The system bootstraps access models by identifying individuals with similar attributes and access patterns, using access control policies to narrow the candidate pool, and aggregating individual models into group models for anomaly detection, with a meta-classifier updating models as data accumulates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical models are built using available training data, then anomaly detection capability is improved, but model reliability deteriorates when data is insufficient
Solution Approach 1:
The patent combines multiple individual access models into a group access model when training data for a specific individual is insufficient. This merging approach allows the system to leverage collective behavioral patterns from multiple users to maintain reliable anomaly detection even when individual data is scarce, directly resolving the contradiction between detection capability and model reliability.
Solution Approach 2:
The system creates a universal group access model that can serve multiple individuals within a group, making the model applicable across different users with similar access patterns. This multi-functional approach ensures that the model remains reliable by drawing from aggregated data while still providing individualized anomaly detection capability.
2Measurement precision
If access models are built for individual users, then detection accuracy is improved, but data requirements increase
Solution Approach 1:
By merging individual access models into a group model, the system reduces the data requirements for each individual user. Instead of needing sufficient training data for every single user, the system aggregates data across the group, thereby maintaining detection accuracy while significantly lowering the data threshold required for model building.
Solution Approach 2:
The system performs preliminary aggregation of access data from multiple users to create a group model before individual models are fully trained. This preliminary action allows the system to have functional models available even when individual data accumulation is incomplete, reducing the overall data requirements needed to achieve operational detection accuracy.
Data Source
AI summary
A plurality of access models comprise patterns of accesses to resources protected by an Access Control System (ACS). The access models are used to classify the accesses to the resources as either normal or anomalous. The system stores attributes for the individuals for whom the access models are available, receives attributes of an individual not associated with an access model, and compares the attributes of the individual not associated with an access model to the attributes of the individuals for whom access models are available. The system further selects an access model for the individual not associated with an access model based on the comparison, and uses the selected access model for the individual not associated with an access model to classify access to the one or more resources protected by the ACS by the individual not associated with an access model as either normal or anomalous.