Distributed AI Model Scoring via Segmented Client-Cloud Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In distributed AI systems, there is a need to securely and efficiently evaluate AI models across multiple devices or systems, while maintaining proprietary control over custom metrics and ensuring security, especially in cloud environments where resource consumption and maintenance can be costly and computationally expensive.
Innovation Solution
A distributed AI scoring system is implemented, where a first computing system determines and sends the result of an AI model to a second computing system, which applies custom scoring algorithms, enabling distributed ownership and security, and allowing for the evaluation of AI model performance and bias testing without introducing non-native code to cloud systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If AI model evaluation is performed in centralized cloud systems, then resource consumption and maintenance costs increase, but security and proprietary control over custom metrics deteriorate
Solution Approach 1:
The system divides the AI evaluation process into separate segments: the AI model and custom scoring algorithms remain on the client's local system, while only the necessary input data is transmitted to the cloud. This segmentation allows secure evaluation without requiring the entire evaluation pipeline to reside in centralized cloud systems, thereby reducing resource consumption and maintaining security.
Solution Approach 2:
The system introduces an intermediary approach where the cloud platform provides infrastructure and platform services, but the actual scoring algorithms and model evaluation logic remain on the client side. This intermediary architecture allows secure data transmission between client and cloud while maintaining proprietary control over critical evaluation components.
2Adaptability or versatility
If custom scoring algorithms are deployed in cloud environments, then evaluation capability is improved, but security and proprietary control are compromised
Solution Approach 1:
The system extracts the custom scoring algorithms and evaluation logic from the cloud environment and places them on the client's local system. This extraction ensures that proprietary evaluation capabilities remain under the client's control while still benefiting from cloud infrastructure for data storage and model hosting, thereby maintaining security without sacrificing adaptability.
Solution Approach 2:
The system implements local quality by running custom scoring algorithms locally on the client's system rather than centrally in the cloud. This allows each client to maintain their own proprietary evaluation metrics and logic without compromising security, while the cloud infrastructure still provides supporting services.
3Reliability
If AI model evaluation is performed locally, then security and proprietary control are maintained, but computational efficiency and resource usage deteriorate
Solution Approach 1:
The system applies partial action by transmitting only the necessary input data to the cloud for evaluation, rather than transmitting the entire model or all processing operations. This partial data transmission approach maintains security by keeping critical evaluation logic local while still benefiting from cloud computational resources for the specific evaluation task, thereby improving productivity without sacrificing proprietary control.
4Reliability
If distributed AI scoring is implemented, then security and proprietary control are improved, but system complexity increases
Solution Approach 1:
The system implements universality by creating a standardized interface and protocol that allows different computing systems to work together in the distributed evaluation architecture. This standardized approach simplifies the overall system complexity while maintaining security and proprietary control, as the same interface can be used across different platforms and configurations.
Data Source
AI summary
An embodiment of the invention may include a method, computer program product, and system for scoring AI models. An embodiment may include a first computing system that determines a result of an AI model based on input data. An embodiment may include a first computing system that sends, to a second computing system, the result of the AI model. An embodiment may include a first computing system that receives, from the second computing system, a scoring result of the AI model.