Systems and methods for distributed machine learning with less vehicle energy and infrastructure cost
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional decentralized learning methods do not consider energy consumption and infrastructure costs for training machine learning models in vehicles, leading to inefficiencies in battery life and infrastructure deployment.
Innovation Solution
Implementing edge encounter scores (EES) and energy-balanced client selection (EBCS) to determine which vehicle trains and aggregates models, minimizing energy consumption and infrastructure costs while maintaining model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If decentralized learning is implemented without considering energy consumption, then model training can be distributed across vehicles, but energy consumption increases leading to reduced battery life
Solution Approach 1:
The patent changes the selection parameter from random or uniform distribution to energy-based probability distribution. Vehicles with lower energy consumption characteristics are selected with higher probability for model training tasks, optimizing the energy efficiency of distributed learning while maintaining training distribution across the vehicle network.
Solution Approach 2:
The patent implements a feedback mechanism where energy consumption data from previous training tasks is collected and used to adjust future task assignments. The system learns from past energy usage patterns and dynamically adjusts which vehicles are selected for training, creating a closed-loop control that continuously optimizes energy efficiency.
2Reliability
If more infrastructure components are deployed to support distributed learning, then model aggregation capability improves, but infrastructure cost increases
Solution Approach 1:
The patent enables vehicles to perform model aggregation among themselves without requiring dedicated central servers or edge infrastructure. The distributed vehicle network self-organizes to aggregate models, eliminating the need for expensive infrastructure deployment while maintaining aggregation capability through peer-to-peer communication.
Solution Approach 2:
The patent makes existing vehicles serve multiple functions: they act as both data collectors and model trainers, and can function as temporary aggregation points. This multi-functionality eliminates the need for dedicated infrastructure components, reducing infrastructure costs while maintaining model aggregation capability through the vehicles themselves.
Data Source
AI summary
A method for updating a machine learning model for vehicles is provided. The method includes calculating a benefit score for each of a pair of vehicles based on an energy for training a machine learning model and a value of training data, selecting one of the pair of vehicles having a higher benefit score as a trainer for training the machine learning model, aggregating, by the trainer, the machine learning models of the pair of vehicles, calculating an edge encounter score for each of the pair of vehicles, selecting one of the pair of vehicles having a higher edge encounter score as a representer, and uploading, by the representer, the aggregated machine learning model to an edge server.