Endpoint AI File Classification for Dynamic Data Sensitivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data classification systems are ineffective for general intellectual property content due to the dynamic nature of sensitivity and the challenges in training models without sharing proprietary information, leading to inaccurate classification and potential data breaches.
Innovation Solution
A distributed, artificial-intelligence-driven file classification system that operates on endpoint devices, analyzing file content and metadata locally without transmitting sensitive data, and aggregates anonymized metadata for centralized analysis to refine classification policies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional classification systems use regular expression matching, then they can classify structured data in narrow domains, but they fail to classify general intellectual property content accurately due to dynamic sensitivity and lack of contextual understanding
Solution Approach 1:
The patent replaces conventional mechanical pattern-matching systems (regular expressions) with an artificial intelligence-based classification system that uses machine learning models. This substitution enables the system to understand contextual nuances, dynamic sensitivity, and semantic meanings of intellectual property content, thereby achieving both high classification accuracy and adaptability to changing conditions.
Solution Approach 2:
The system dynamically adjusts classification parameters and model behavior based on changing conditions such as public disclosures and sensitivity levels. The AI model can learn from new data and update its understanding of what constitutes sensitive intellectual property information, allowing the classification accuracy to be maintained despite changing contextual parameters.
2Measurement precision
If enterprises train AI models using their intellectual property data, then classification accuracy improves, but proprietary information is exposed to training environments and potential data breaches occur
Solution Approach 1:
The patent introduces a centralized AI training platform as an intermediary between enterprise data and the classification models. This intermediary infrastructure enables model training using enterprise data while implementing security measures to prevent direct exposure of proprietary information to external training environments. The platform acts as a secure mediator that processes data locally and shares only necessary information for model improvement.
Solution Approach 2:
Instead of exposing original proprietary data to external training environments, the system creates and uses synthetic copies or representations of the data for training purposes. This allows the AI models to learn from the patterns and characteristics of intellectual property content without actually exposing the sensitive original data to potentially vulnerable training infrastructure.
3Measurement precision
If centralized AI training platforms aggregate data from multiple sources, then model accuracy improves, but data transmission and processing complexity increase
Solution Approach 1:
The patent divides the data aggregation and processing function into segmented components distributed across multiple sources and processing stages. Rather than a single monolithic centralized platform, the system segments data collection, anonymization, aggregation, and model training into separate functional modules that can operate independently and scale as needed, reducing overall system complexity.
Solution Approach 2:
The system extracts and removes sensitive or identifying information from the data before aggregation and training. By taking out proprietary information, personally identifiable data, and other sensitive elements from the raw data streams, the system reduces the complexity of secure processing while still maintaining the useful patterns and features needed for accurate classification models.
Data Source
AI summary
Various aspects relate to mechanisms for data object classification in connection with a memory and a processor. At an endpoint device, a classification of a data object is determined based on output from a machine learning model configured to take as input contents and metadata of the data object, wherein the classification comprises a confidence score. It is determined whether the data object requires additional review, based on the confidence score. A data object hash is computed based on the contents and the metadata of the data object. An internal structure of the machine learning model is updated based on the additional review, the data object hash, and subsequent operation of the endpoint device.