On-Premises Deep Learning Classifier Training for Sensitive Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data loss prevention (DLP) technologies face challenges in accurately detecting sensitive information in images due to the high computational demands of optical character recognition (OCR) and the need for large numbers of high-quality labeled images, which are difficult to acquire, especially when privacy concerns prevent customers from sharing their data with DLP providers.
Innovation Solution
A Train Your Own Classifier (TYOC) mechanism using a customized deep learning (DL) stack classifier is deployed on-premises, allowing organizations to train their own classifiers using Docker containers, converting images to numeric features without sharing the original data, thus ensuring data privacy and security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR-based DLP technologies are used to detect sensitive information in images, then detection capability is provided, but computational costs are high and detection accuracy is insufficient
Solution Approach 1:
The patent replaces traditional OCR-based mechanical text recognition systems with deep learning-based neural network models. The DL stack classifier uses learned features from training data to directly classify sensitive information in images, substituting the step-by-step mechanical OCR process with a more efficient neural network approach that achieves higher accuracy while reducing computational overhead through feature extraction and pattern recognition.
Solution Approach 2:
The patent transforms images into numeric feature representations through the deep learning stack, changing the parameter space from raw pixel data to extracted features. This parameter transformation enables more efficient processing by the classifier, reducing the computational dimensionality while preserving the essential characteristics needed for accurate sensitive information detection.
2Measurement precision
If customers share their image data with DLP providers for training, then classifier accuracy can be improved, but data privacy and security concerns arise
Solution Approach 1:
The patent introduces an on-premises Docker container as an intermediary between the customer's data and the cloud-based DLP service. The DL stack classifier is deployed locally in the container, allowing training and operation to occur within the customer's secure environment. This intermediary arrangement enables classifier improvement through local training while maintaining data privacy, as the sensitive images never leave the customer's premises.
Solution Approach 2:
The patent enables customers to train their own classifiers using their own data within their own infrastructure. The on-premises DL stack allows organizations to perform self-service training by providing their labeled images to the local model, improving accuracy for their specific use cases without sharing data externally. This self-service approach empowers customers to maintain both high accuracy and data sovereignty.
3Object-affected harmful factors
If deep learning classifiers are deployed on-premises using Docker containers, then data privacy is protected, but system complexity increases
Solution Approach 1:
The patent leverages Docker containers as a universal platform that provides multiple functions: isolation of the DL stack, management of dependencies, portability across environments, and simplified deployment. This multi-functional container approach reduces the overall system complexity by consolidating infrastructure requirements into a single, well-understood technology that handles security, resource management, and deployment automation.
Solution Approach 2:
The patent uses Docker containerization to create a portable, copyable instance of the DL stack that can be deployed on-premises. The container encapsulates the entire deep learning environment with all necessary dependencies, allowing the complex DL system to be treated as a simple, self-contained unit that can be copied and deployed without requiring complex infrastructure setup or configuration management.
Data Source
AI summary
Disclosed is a method of a classifier Machine Learning (ML) training platform to train a custom classifier without accessing organization sensitive data in images, referred to as organization sensitive documents, and protecting against exfiltration of the image-borne organization sensitive documents. The method includes receiving, from an organization, organization-specific examples including non-invertible feature maps extracted from organization-sensitive documents and ground truth labels without receiving the organization-sensitive documents. The method includes using the received organization-specific examples to train a customer-specific Machine Learning (ML) stack classifier using the non-invertible feature maps and the ground truth labels. The method includes sending the customer-specific DL stack classifier to the organization.