Edge Computing Model Selection via Core Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Edge computing devices with limited resources struggle to efficiently perform machine learning-based computations due to their inability to locally train accurate models, leading to suboptimal performance and resource constraints.
Innovation Solution
A method where a core computing system continually trains multiple machine learning models based on data from edge devices, monitoring network and device states to select and update models in edge devices, allowing them to switch between models based on resource availability and performance metrics, without requiring additional resources from the edge devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained locally on edge computing devices, then model accuracy can be improved, but device resource constraints (power, computation, memory) prevent effective training
Solution Approach 1:
A core computing system acts as an intermediary between data sources and edge devices. The core system collects data from edge devices, trains multiple ML models with different accuracy-resource profiles, and distributes appropriate models to edge devices based on their current resource availability and performance requirements, thereby achieving accurate modeling without burdening edge device resources
Solution Approach 2:
The system changes the parameter of model accuracy by maintaining multiple versions of ML models with different complexity and accuracy levels. Edge devices can switch between these model versions depending on their current resource state, allowing them to achieve high accuracy when resources permit while conserving energy when resources are constrained
2Adaptability or versatility
If multiple machine learning models are trained to provide flexibility and accuracy, then model performance can be optimized, but training resources and computational overhead increase
Solution Approach 1:
Multiple ML models are trained in advance at the core computing system before being deployed to edge devices. This preliminary training phase allows the system to explore different model architectures and hyperparameters, creating a library of pre-trained models that can be selectively deployed based on edge device needs without performing complex training operations at the edge
Solution Approach 2:
The system segments the ML model training and deployment process into distinct phases: data collection at edge, model training at core, and model selection/deployment back to edge. This segmentation allows complex training operations to be performed centrally while keeping edge device complexity low, maintaining adaptability through multiple model versions
3Measurement precision
If machine learning computations are performed at the core system, then model accuracy can be maintained, but network bandwidth consumption increases
Solution Approach 1:
The system extracts only the essential model parameters and weights from the training process at the core system, transmitting only these compressed representations to edge devices rather than transmitting raw training data or full model computations. This extraction approach maintains model accuracy while significantly reducing network bandwidth consumption
4Reliability
If edge devices continuously update their models based on local data, then model relevance to local conditions can be improved, but computational resources and energy consumption at edge devices increase
Solution Approach 1:
The core computing system serves as an intermediary that collects local data from edge devices, performs the computationally intensive model training and updating, then distributes updated models back to edge devices. This approach maintains model relevance to local conditions while avoiding continuous local training computations at resource-constrained edge devices
Data Source
AI summary
Optimizing a network comprising a core computing system (CCS) and a set of edge computing devices (ECDs), wherein each of the ECDs locally performs computations based on a trained machine learning (ML) model. A plurality of ML models are continually trained at the CCS, concurrently, based on data collected from the ECDs. One or more states of the network and/or components thereof are monitored. The monitored states are relied upon to decide (when) to change a trained ML model as currently used by any of the ECDs to perform said computations. It may be decided to change the model used by a given one of the ECDs to perform ML-based computations. One of the models as trained at the CCS is selected (based on the monitored states) and corresponding parameters are sent to this ECD. The latter can resume computations according to a trained model.