Federated Learning Client Selection and Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated learning, client devices with poor data quality or lower data volume, along with varying hardware specifications and network speeds, lead to reduced learning efficiency and delayed training times due to unbalanced data distribution and synchronous calculation delays.
Innovation Solution
A federated learning method and system that prioritizes client devices based on importance parameters and groups them by performance parameters, allowing target devices to perform iterations and generate trained models, which are then used to update a global model, ensuring convergence within a default range and number of iterations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If client devices with poor data quality or less data volume are selected to participate in training with the same probability, then the selection process is simple and fair, but the learning efficiency of model training is reduced
Solution Approach 1:
The patent changes the selection parameter from uniform random selection to selection based on data quality metrics and data volume. Each client device is evaluated using importance parameters (data quality, data volume) and performance parameters (hardware specifications, network speed), and selection probability is adjusted according to these parameters, thereby improving learning efficiency while maintaining a systematic selection process
2Adaptability or versatility
If client devices with different hardware specifications and network speeds are selected, then more diverse devices can participate in training, but the overall training time is delayed due to waiting for slow devices
Solution Approach 1:
The patent applies local quality by differentiating device selection and grouping based on performance parameters. Devices are evaluated individually on hardware specifications and network speed, then grouped into training groups where devices with similar performance characteristics are trained together. This allows each group to progress at its own pace without being bottlenecked by slower devices in mixed-performance groups
Solution Approach 2:
The patent segments the client devices into different training groups based on performance parameters. Instead of forming a single homogeneous training group, devices are divided into multiple groups with similar performance characteristics, allowing parallel training progression and reducing overall training time by eliminating waiting periods for slower devices
3Stability of the object's composition
If all selected client devices must complete training synchronously, then the global model can be updated consistently, but fast devices must wait for slow devices, reducing overall training efficiency
Solution Approach 1:
The patent segments devices into performance-based training groups that can progress through training iterations independently. Each training group completes its training and submits models to the central device, which then aggregates them to update the global model. This segmentation allows fast devices to complete training without waiting for slow devices while maintaining model consistency through centralized aggregation
Solution Approach 2:
The patent performs preliminary grouping of devices based on performance parameters before training begins. This preliminary action ensures that devices are pre-configured into appropriate training groups, allowing them to proceed with training at optimized speeds without synchronization delays during the training process itself
Data Source
AI summary
A federated learning method includes: providing importance parameters and performance parameters by client devices respectively to a central device, performing a training procedure by the central device, wherein the training procedure includes: selecting target devices from the client devices according to a priority order associated with the importance parameters, dividing the target devices into training groups according to a similarity of the performance parameters, notifying the target devices to perform iterations according to the training groups respectively to generate trained models, transmitting the trained models to the central device, and updating a global model based on the trained models, performing the training procedure again or outputting the global model to the client devices based on a convergence value of the global model and the number of times of performing the training procedure.