Federated AI Training via Generative Adversarial Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI training methods for edge-based applications face challenges in balancing network performance with user privacy and device resource preservation, as they either rely on generic data that lacks user specificity or send sensitive user data to central servers, compromising privacy.
Innovation Solution
A method employing a generative adversarial network to train a federated classifier using edge-trained models from multiple devices, generating data samples without exchanging private user data, and merging these models into a single AI model for deployment back to edge devices, preserving user specificity and privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If user data is sent to a central server for model training, then the model incorporates all user information and achieves high network performance, but user privacy is compromised
Solution Approach 1:
The patent extracts only the trained model parameters from edge devices and sends them to the central server for aggregation, rather than transmitting raw user data. This extraction approach allows the system to achieve high network performance by incorporating all user information while preserving user privacy, as the sensitive raw data remains localized on edge devices.
Solution Approach 2:
The patent introduces an intermediary mechanism where trained models from multiple edge devices serve as intermediaries to transfer information to the central server. Instead of direct data transmission, the intermediary trained models convey user-specific patterns without exposing raw user data, thus resolving the contradiction between network performance and privacy preservation.
2Productivity
If a centralized database is used to train deep neural networks, then training can be performed efficiently, but user specificity is lost in the generic model
Solution Approach 1:
The patent applies local quality by training individual models at each edge device with local user-specific data, creating models with localized characteristics. Each edge device maintains its own trained model that reflects user-specific patterns, and these localized models are then aggregated at the central server to create a federated model that preserves user specificity while maintaining training efficiency.
Solution Approach 2:
The patent segments the training process into distributed local training at edge devices and centralized aggregation at the server. This segmentation allows each segment (edge device) to train independently with local data, preserving user specificity, while the overall system maintains training efficiency through parallel processing and reduced data transmission requirements.
3Productivity
If all user data is transmitted to the cloud for training, then a comprehensive model can be created, but bandwidth consumption and device power are excessive
Solution Approach 1:
The patent extracts only the essential trained model parameters from edge devices and transmits them to the central server, rather than transmitting all raw user data. This extraction approach creates a comprehensive model at the server while significantly reducing bandwidth consumption and device power requirements, as only compressed model parameters are transmitted.
Solution Approach 2:
The patent uses copying by creating replicated trained models at each edge device and transmitting these copies to the central server for aggregation. This copying mechanism allows the system to build a comprehensive model from multiple sources while minimizing data transmission, as only the copied model parameters (not the original large-scale raw data) are transmitted across the network.
Data Source
AI summary
A computer-implemented method is presented for constructing a trained model for a plurality of edge classifiers in a network having a federated classifier, a generator, and a discriminator. The method includes obtaining edge trained models from the plurality of edge devices, each edge trained model being trained independently with data from private data of each edge, training the generator model and discriminator model by employing the edge trained models and an unlabeled set of data by employing a generative adversarial training procedure, generating data samples by the trained generator model, training the federated classifier with the data samples from the generator model, and deploying the trained model back to the plurality of edge devices.