Balanced Client Selection for Non-IID Vehicle Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Centralized machine learning in vehicles is impractical due to high communication overhead, infrastructure costs, and privacy concerns, while decentralized and federated learning face challenges from non-IID data causing accuracy degradation and inefficiencies.
Innovation Solution
A balanced client selection system for decentralized machine learning (BCS-DL) that performs real-time model aggregation and weight allocation without a global perspective, using training contribution estimation (TCE) and model weight computation (MWC) to mitigate non-IID data biases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If centralized machine learning is used, then model training can be performed with shared global models, but communication overhead and infrastructure costs increase significantly
Solution Approach 1:
The patent segments the centralized training process into distributed local training processes across multiple vehicles. Each vehicle trains its own local model using its local data, eliminating the need for all data to be uploaded to a central server. This segmentation directly reduces communication overhead while maintaining training reliability through local computation.
Solution Approach 2:
Each vehicle performs self-service by training its own local model using its local data without requiring centralized coordination. The vehicle independently calculates class-wide data contribution estimations and aggregates models with neighboring vehicles, making the system autonomous and eliminating heavy infrastructure dependencies.
2Loss of energy
If decentralized machine learning is used, then communication overhead is reduced, but non-IID data causes accuracy degradation
Solution Approach 1:
The patent applies local quality by calculating class-wide data contribution estimations specific to each vehicle's local data distribution. Each vehicle determines its own contribution weight based on its local non-IID data characteristics, allowing the system to adapt to local data qualities while maintaining overall model accuracy through weighted aggregation.
Solution Approach 2:
The system changes the aggregation parameters dynamically by calculating contribution weights based on class-wide data statistics. Instead of uniform aggregation, the system adjusts aggregation weights according to each vehicle's local data contribution, transforming the aggregation process to accommodate non-IID data distributions while preserving accuracy.
3Loss of energy
If real-time client selection is performed without global perspective, then infrastructure costs are reduced, but model aggregation becomes more complex
Solution Approach 1:
The patent extracts the global coordination function from the aggregation process. Instead of requiring a central controller to manage all aggregations, each vehicle independently calculates its own data contribution and performs local aggregation with neighboring vehicles. This extraction of global perspective reduces infrastructure costs while managing complexity through localized operations.
Solution Approach 2:
The system introduces class-wide data contribution estimation as an intermediary mechanism that simplifies the aggregation process. This intermediary calculation provides a standardized metric for weighing different vehicles' contributions, transforming the complex problem of real-time client selection into a simpler weighted aggregation task that can be performed locally without global coordination.
Data Source
AI summary
Systems and methods for implementing a balanced client selection for decentralized machine learning (BCS-DL) that can be utilized in both decentralized machine learning and hybrid machine learning for vehicle environments are described. For example, a vehicle can include a processor device training a machine learning model using local data, and a controller device performing balanced client selection in real-time to communicate the local machine learning model with a connected vehicle for decentralized machine learning. Hybrid machine learning combines aspects from federated learning and decentralized machine learning approaches. The disclosed BCS-DL system is designed to execute balanced client selection in real-time for vehicles that are acting as clients in a hybrid machine learning infrastructure. The balanced client selection also calculates a training contribution estimation (TCE) and a model weight computation (MWC) to mitigate imbalance in the data distribution incurred by non-Independent, Identically Distributed (non-IID data) related to decentralized and hybrid machine learning.