Branched Neural Object Detection With Credential-Based Access Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection systems fail to adequately differentiate user rights, allowing unauthorized users to access sensitive information by detecting secret objects as 'blind spots', necessitating separate systems for each user combination.
Innovation Solution
A branched neural network system with credential management that selectively activates branches based on user credentials, ensuring only authorized users access specific information by disabling irrelevant branches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional object detection is used to filter secret objects for unauthorized users, then access control is implemented, but detection blind spots are created that reveal the filtering behavior
Solution Approach 1:
The neural network is divided into multiple independent branches, each capable of detecting different object classes. This segmentation allows the system to process detections from all branches and then filter results at the output stage, preventing blind spots while maintaining access control. Each branch operates independently to detect its target class without being influenced by user credentials.
Solution Approach 2:
A credential manager component acts as an intermediary between the branched neural network and the output. It receives detection results from all branches and applies access control policies to determine which detections are visible to each user, thereby preventing blind spots while enforcing security requirements.
2Reliability
If separate object detection systems are replicated for each user combination, then user-specific access control is achieved, but system complexity and resource requirements increase significantly
Solution Approach 1:
A single branched neural network system serves multiple user types and access control requirements simultaneously. The system processes sensor data once and generates detections for all object classes, then the credential manager filters results based on user credentials. This eliminates the need to replicate entire detection systems for different user combinations.
Solution Approach 2:
Multiple detection functions for different object classes are merged into a single branched neural network architecture. All branches process the same sensor data in parallel and their results are combined, with access control applied at the output stage. This consolidation reduces system complexity while maintaining user-specific access control.
3Reliability
If the entire object detection system is replicated for each user class, then secure access control is achieved, but computational resources and processing time increase
Solution Approach 1:
The detection system is segmented into independent branches that can be selectively activated based on user credentials. The credential manager enables only the necessary branches for each user, avoiding unnecessary computational work while maintaining security. This selective activation improves processing efficiency compared to running all branches for all users.
Solution Approach 2:
The system performs partial detection by activating only the subset of branches necessary for each user's access level. Instead of running the complete detection pipeline for all possible object classes, the credential manager enables only relevant branches, reducing computational overhead while ensuring secure access control for authorized detections.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus for detecting objects in sensor data (30) by users is disclosed, wherein user credentials (60) are assigned to each user. The apparatus comprising: an input engine (110) configured to receive as input the sensor data (30); a branched neural network machine (120) with multiple branches (121, 122, 123, 124, 221, 222, 223), the multiple branches (121, 122, 123, 124, 221, 222, 223) being configured to operate at least temporarily in parallel and each branch being specialized for branch-specific input data received from the input engine (110); a credential manager (130) configured to selectively activate branches of the multiple branches (121, 122, 123, 124, 221, 222, 223) based on the user credentials (60); and a combiner (140) configured to combine outputs of the branched neural network machine (120) and to output (150) indicating, dependent on the user credentials, detected objects in the sensor data (30)