Edge ML Feature Selection via Central Contextual Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing dimensionality reduction techniques in machine learning models deployed at the edge of a network often ignore relevant and causally important data, leading to suboptimal performance due to the lack of consideration for contextual information.
Innovation Solution
A method is introduced where a central entity selects a set of features for a machine learning model based on available data, features, and contextual information associated with the network, using processing circuitry and memory to communicate and process this information, thereby improving the model's performance by incorporating previously irrelevant data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If dimensionality reduction techniques are used to reduce data at the edge, then compute resources and storage resources are saved, but relevant and causally important data may be ignored or excluded
Solution Approach 1:
The patent applies preliminary action by performing dimensionality reduction and feature selection at a central entity before deploying the machine learning model to the edge entity. This allows the central entity to identify and retain causally important features using full contextual information, ensuring that only truly relevant features are reduced at the edge, thus saving compute resources without losing important data.
Solution Approach 2:
The patent introduces a central entity as an intermediary between the data source and the edge machine learning model. This intermediary performs feature selection and dimensionality reduction using comprehensive contextual information, then transmits the selected features to the edge entity. This mediator ensures that causally important data is preserved while still achieving data reduction for efficient edge computing.
2Quantity of substance
If dimensionality reduction techniques are used to reduce data at the edge, then storage resources and transport resources are saved, but the ML model may ignore data that could improve performance
Solution Approach 1:
The system performs preliminary feature selection at the central entity using complete contextual information before deploying the model to the edge. This advance processing ensures that only the most relevant features are selected for transmission to the edge, reducing data volume while maintaining model reliability by preserving causally important features.
Solution Approach 2:
The patent implements a feedback mechanism where the central entity receives information about data availability and model performance, then adjusts feature selection accordingly. This feedback loop ensures that features critical for model performance are identified and retained, while still achieving data reduction for efficient storage and transport at the edge.
3Loss of time
If reduced data is used for training and prediction at the edge, then response time constraints are met, but overfitting may be avoided less effectively
Solution Approach 1:
The system performs preliminary dimensionality reduction and feature selection at the central entity using full contextual information before deploying to the edge. This advance processing creates an optimized feature set that maintains model generalization capabilities while reducing the data volume processed at the edge, thus meeting response time constraints without sacrificing reliability.
Solution Approach 2:
The central entity acts as an intermediary that performs feature selection using comprehensive data, then transmits the optimized feature set to the edge entity. This mediator ensures that the reduced feature set maintains good generalization properties by preserving causally important features, while the edge entity can process the reduced data quickly to meet response time requirements.
4Reliability
If the ML model uses more data locally available at the edge, then better performance can be achieved, but compute resources and processing power are consumed
Solution Approach 1:
The system performs preliminary feature selection at the central entity using complete contextual information and full data sets, identifying the most causally important features before deployment. This advance processing allows the edge model to achieve good performance using only the selected features, avoiding the need to process all locally available data and thus conserving compute resources at the edge.
Solution Approach 2:
The patent extracts only the most relevant and causally important features from the complete data set using dimensionality reduction techniques at the central entity. This extraction process creates a streamlined feature set that can be processed efficiently at the edge, achieving good model performance while minimizing compute resource consumption by excluding irrelevant features.
Data Source
AI summary
There is provided a method performed by a central entity of a network. A first set of features is selected for a machine learning model to take into account when analysing data. The machine learning model is to be deployed at an edge entity of the network. The selection is based on first information indicative of data that is available for the machine learning model to analyse, second information indicative of features that are available for the machine learning model to take into account when analysing data, and contextual information associated with the network.