Edge Device Federated Learning Model Utility Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional federated learning on edge devices lacks coordination between competing services, leading to resource constraints and inefficient training of multiple machine learning models, as existing methods do not account for the edge device's resource limitations.
Innovation Solution
An apparatus and method that rank FL models based on utility, selecting high-utility models first and allocating training samples efficiently to maximize utility gain within a given cost budget, using a combination of high-loss and low-loss samples to optimize training on edge devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the edge device participates in multiple FL training tasks from competing services, then the service coverage and model diversity are improved, but the resource constraints (energy, computation, memory) are exceeded
Solution Approach 1:
The system dynamically changes training parameters including sample selection ratios (high-loss vs low-loss samples), training iterations, and model architecture configurations based on available energy budget and device state, allowing adaptation between service coverage and energy consumption
Solution Approach 2:
The edge device performs partial training on selected models rather than complete training on all available models, using a subset of training samples and limited iterations to achieve useful model performance while staying within energy constraints
2Productivity
If the edge device trains all available FL models from competing services, then the model utility and service participation are maximized, but the training time and computational cost exceed available resources
Solution Approach 1:
The system adjusts training parameters such as number of epochs, batch size, and sample selection based on time constraints and priority rankings, dynamically optimizing the trade-off between model utility and training time
Solution Approach 2:
The training process is segmented into priority levels where high-utility models receive full training resources while lower-utility models receive reduced training or are skipped entirely, allowing time-constrained selective training
3Measurement precision
If the edge device uses more training samples for each FL model, then the model accuracy and performance are improved, but the aggregate cost budget is exceeded
Solution Approach 1:
The system changes the quantity and quality of training samples based on model priority and budget constraints, using more samples for high-utility models and fewer or no samples for low-utility models, optimizing the accuracy-cost trade-off
Solution Approach 2:
Different quality levels of training samples (high-loss vs low-loss) are selectively applied to different models based on their utility rankings, with higher quality samples allocated to higher priority models
4Device complexity
If the FL service uses server-centric client selection, then the model training coordination is simplified, but the resource constraints of edge devices are not considered leading to inefficient training
Solution Approach 1:
An intermediary selection mechanism is introduced that considers both server objectives and device constraints, acting as a mediator between the FL service provider and edge device resources to achieve efficient model selection and training allocation
Data Source
AI summary
An apparatus, method and computer program is described comprising receiving, at an edge device from one or more federated learning (FL) services, one or more FL machine learning (ML) models and capabilities information associated with each FL ML model, computing the utility of each FL ML model based a quality of available training samples at the edge device for training the corresponding FL ML models and said capabilities information, ranking the FL ML models in a descending order of utility of the FL ML models; and performing training of each of the corresponding FL ML models of each FL service in a descending order of utility until a remaining available cost budget of a total available cost budget of the edge device expires.