Vehicle Selection by Edge Encounter Score for Decentralized Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional decentralized learning methods for connected vehicles lack a clear mechanism for selecting a vehicle to collect and aggregate machine learning models from a pair of vehicles, leading to inefficiencies in communication and infrastructure requirements.
Innovation Solution
A method and system for selecting a vehicle based on an 'edge encounter score' calculated from movement momentum and direction to edge servers, allowing the selected vehicle to aggregate models and upload them to an edge server, thereby reducing communication and computational demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If decentralized learning is implemented without a central server, then infrastructure costs are reduced, but it becomes unclear which vehicle should collect and aggregate models leading to communication inefficiency
Solution Approach 1:
The system enables vehicles to autonomously determine their role in model aggregation through self-calculated edge encounter scores based on their own movement momentum and direction to edge servers, eliminating the need for central server coordination while maintaining clear aggregation protocols
Solution Approach 2:
The patent introduces edge encounter score as a new parameter that dynamically changes based on vehicle movement momentum and direction to edge servers, providing a quantitative basis for role assignment in decentralized model aggregation
2Reliability
If federated learning is used with central server aggregation, then model convergence is maintained, but communication requirements among vehicles, edge servers, and central server increase significantly
Solution Approach 1:
The patent extracts the central server aggregation function and redistributes it to vehicles with higher edge encounter scores, eliminating the need for all vehicles to communicate with both edge servers and central server, thereby reducing overall communication requirements while preserving model convergence through structured aggregation protocols
3Device complexity
If all vehicles upload models directly to edge servers, then model aggregation is simplified, but communication and computational demands on edge servers increase
Solution Approach 1:
The patent segments the model aggregation process by introducing intermediate vehicles that perform local aggregation before uploading to edge servers. This divides the computational burden from edge servers and reduces the quantity and size of models that edge servers must process, while maintaining aggregation effectiveness through hierarchical structure
Data Source
AI summary
A method for updating a machine learning model for vehicles is provided. The method includes obtaining an edge encounter score for each of a pair of vehicles calculated based on a movement momentum of each of the pair of vehicles and a direction from a location of each of the pair of vehicles to each of one or more edge servers, selecting one of the pair of vehicles having a higher edge encounter score than other vehicle, aggregating a machine learning model of the selected vehicle and a machine learning model of the other vehicle of the pair of vehicles, uploading the aggregated machine learning model to one of the one or more edge servers, and operating the selected vehicle based on the aggregated machine learning model.