AI Service Platform Controller for Privacy-Preserving Federated-Split Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence (AI) technologies face challenges in providing privacy-preserving deep learning with decentralized data, particularly in federated learning (FL) and split learning (SL), where information leakage and scalability issues hinder effective model training and user privacy is not adequately protected.
Innovation Solution
A system and method that utilizes a platform controller to manage AI services, enabling decentralized data management through a multi-level learning framework combining federated and split learning, with a coordinator facilitating anonymous interactions between devices and servers, ensuring k-anonymity and differential privacy by selecting optimal cut layers for balanced learning overheads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If federated learning is used to train deep learning models with decentralized data, then data privacy is protected, but model accuracy deteriorates due to non-IID local data
Solution Approach 1:
The system segments the learning process into two distinct levels: federated learning at the top level for global model aggregation, and split learning at the bottom level for local model training. This segmentation allows each level to operate optimally - FL protects privacy while SL maintains accuracy by using local data without requiring it to leave the device.
Solution Approach 2:
The patent merges federated learning and split learning into a unified multi-level framework. The FL component handles global coordination and model aggregation, while the SL component handles local computation and data processing. This combination leverages the privacy protection strengths of FL and the accuracy preservation strengths of SL.
2Measurement precision
If split learning is used to train deep learning models with local data, then model accuracy is maintained, but scalability deteriorates due to sequential learning requirements
Solution Approach 1:
The system adds a temporal and organizational dimension to split learning by structuring it as the bottom level of a multi-level framework. This allows multiple split learning processes to occur in parallel across different client devices, with results aggregated at the federated learning level, thereby improving scalability while maintaining accuracy.
3Reliability
If a multi-level learning framework combining federated and split learning is implemented, then privacy preservation and model accuracy are improved, but system complexity increases
Solution Approach 1:
The complex multi-level learning system is segmented into clearly defined modules: a platform controller for management, coordinators for facilitating interactions, and distinct top (federated) and bottom (split) learning levels. This segmentation makes the complex system more manageable and easier to implement.
Solution Approach 2:
The patent introduces coordinators as intermediary components that facilitate anonymous interactions between devices and servers. These coordinators handle the complexity of multi-level learning coordination, privacy management, and data routing, thereby reducing the burden on individual devices and simplifying the overall system architecture.
4Reliability
If coordinators facilitate anonymous interactions to ensure k-anonymity, then user privacy is protected, but communication overhead increases
Solution Approach 1:
The system performs preliminary actions by having coordinators pre-establish anonymous communication channels and pre-manage device-group associations before actual learning tasks begin. This preliminary setup reduces communication overhead during the actual learning process, as devices can interact through pre-configured anonymous channels without repeated identity verification or channel establishment.
Data Source
AI summary
There is provided a system including a platform controller for managing artificial intelligence services, wherein the system includes a processor coupled with a memory, having stored thereon instructions. The instructions, when executed by the processor, configure the platform controller to receive an artificial intelligence (AI) service registration request from an AI controller controlling the AI service, the AI service registration request including information indicative of locations of the AI service and transmit an AI service registration response to the AI controller, the AI service registration response including routing information at least in part specifying how to reach a coordinator associated with the AI service, the coordinator corresponding to a location of the AI service and transmit a notification indicative of availability of the AI service to a device. When a request for access to the AI service is received from the device, the platform controller is configured to transmit a response to the device, wherein the response is indicative of whether the request is accepted.