Edge Device Scheduling for Federated Learning Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In traditional machine learning technologies, the collection and training of data on terminal devices pose significant privacy risks, whereas federated learning retains data locally, but inefficiencies in scheduling edge devices for training can lead to reduced federated learning efficiency due to performance differences and resource limitations.
Innovation Solution
A distributed machine learning method that acquires an optimal scheduling queue for edge devices based on delay information, balancing computation and communication delays to improve scheduling efficiency, allowing for rational and efficient training of machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If edge devices are scheduled for federated learning training, then distributed machine learning can be performed with privacy protection, but performance differences and resource limitations among edge devices reduce training efficiency
Solution Approach 1:
The patent changes the scheduling parameter from simple round-robin or random selection to delay-aware scheduling. By monitoring and adjusting scheduling decisions based on real-time delay information (computation delay and communication delay) of each edge device, the system optimizes training efficiency while maintaining privacy protection through federated learning architecture.
Solution Approach 2:
The patent implements a feedback mechanism where the server collects delay information from edge devices after each training round and uses this feedback to dynamically adjust the scheduling queue for the next round. This closed-loop control ensures that devices with better performance are prioritized, improving overall federated learning efficiency while preserving data privacy.
2Adaptability or versatility
If edge devices with different performance levels are scheduled together, then more devices can participate in training, but the overall training time is extended due to waiting for slower devices
Solution Approach 1:
The patent performs preliminary assessment of edge device performance by collecting delay information before constructing the scheduling queue. By pre-evaluating computation delay and communication delay of each device, the system can proactively arrange an optimized scheduling order that minimizes waiting time, allowing faster devices to contribute more without extending overall training time.
Solution Approach 2:
The patent introduces asymmetric scheduling where edge devices are not treated equally but are assigned different priorities based on their performance characteristics. Devices with lower delays are scheduled earlier or given more training rounds, creating an asymmetric distribution of training opportunities that optimizes overall system efficiency while maintaining flexibility in device participation.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The present disclosure provides a distributed machine learning method and system, a server, a device and a storage medium, and relates to the field of artificial intelligence technologies, such as machine learning technologies, or the like. An implementation includes: acquiring, based on delay information, an optimal scheduling queue of a plurality of edge devices participating in training; and scheduling each edge device of the plurality of edge devices to train a machine learning model based on the optimal scheduling queue of the plurality of edge devices. The present disclosure may effectively improve a distributed machine learning efficiency.