Dynamic Server-Vehicle Load Balancing for Machine Learning Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing machine learning system for vehicle control faces high processing loads on servers during training, which can lead to a significant drop in processing ability or even shutdowns, especially when training multiple vehicles simultaneously.
Innovation Solution
A method where vehicles and servers dynamically adjust the processing load by determining the optimal distribution of training tasks based on current server load, with vehicles taking on training when the load is high and servers handling it when the load is low, and adjusting based on the type of machine learning model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the server performs training of machine learning models for multiple vehicles simultaneously, then the training capability and model accuracy are improved, but the processing load on the server becomes excessively high
Solution Approach 1:
The training task is segmented and distributed between the server and individual vehicles. The server performs training for vehicles with low processing loads, while vehicles with high processing loads perform training locally. This segmentation reduces the集中 processing load on the server while maintaining overall training capability across the system.
Solution Approach 2:
The system dynamically determines the processing amount or processing ratio of training at the server based on the current processing load. When the server's processing load is high, the training processing ratio at the server is reduced; when the load is low, the ratio is increased. This dynamic adjustment allows the system to adapt to changing conditions and prevent server overload.
2Reliability
If the server reduces training processing to maintain low processing load, then server stability is improved, but the training efficiency and model update speed decrease
Solution Approach 1:
Vehicles are empowered to perform training independently when the server's processing load is high. Each vehicle uses its own computing resources to train its machine learning model locally, reducing dependency on the server during high-load periods. This self-service capability ensures training continues without compromising server stability.
Solution Approach 2:
The system dynamically adjusts the distribution of training tasks between server and vehicles based on real-time processing load conditions. This dynamic balancing ensures that training efficiency is maintained by shifting work to vehicles when appropriate, while server stability is preserved by reducing its processing burden during high-load periods.
Data Source
AI summary
A machine learning method performs training of values of model parameters forming a machine learning model, and is performed in a machine learning system including a vehicle having the machine learning model and a server able to communicate with the vehicle. The machine learning method includes: detecting, by the server, a current processing load of the server; determining, by the server, processing amounts of training respectively performed by the server and the vehicle, based on the processing load; performing, by the server, training of the values of the model parameters in accordance with the determined processing amount in the server; and performing, by the vehicle, training of the values of the model parameters in accordance with the determined processing amount in the vehicle. The server decreases a processing amount at the server when the processing load is relatively high compared to when it is relatively low.