Dynamic Feature Selection for Unreliable Edge Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous mobile edge devices face performance issues due to unreliable sensor data, which complicates real-time predictions and decisions in environments like warehouses and factories, where multiple sensors with varying reliability are used.
Innovation Solution
A method for automatically selecting the best subset of features for predictions based on feature health scores and lazy feature importance measures, allowing for adaptable and reliable predictions in edge environments without requiring the selection of an optimal machine learning model for all edge devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If data from multiple sensors is used for predictions, then the coverage and information availability are improved, but the reliability of predictions deteriorates due to unreliable sensor data
Solution Approach 1:
The system dynamically changes the parameters of feature selection by computing importance scores at inference time based on current sensor readings and data point characteristics. This allows the system to adapt which features are used for prediction based on their current reliability and relevance, rather than using a fixed set of features from all sensors.
Solution Approach 2:
The feature selection process is made dynamic by calculating lazy feature importance measures at inference time rather than pre-computing them. The system adapts the subset of features used for prediction based on the specific data point being evaluated and current sensor reliability, enabling real-time adaptation to changing conditions.
2Measurement precision
If all available features are used for prediction, then the prediction accuracy is improved, but the computational complexity and processing time increase
Solution Approach 1:
The system applies partial action by selecting only the most relevant subset of features for each prediction based on computed importance scores. Instead of processing all available features, the system identifies and uses only the necessary portion of features that contribute most to prediction accuracy for the specific data point.
Solution Approach 2:
The system changes the parameter of feature subset size dynamically based on the specific prediction task and data point characteristics. By adjusting which features are included in the prediction based on their computed importance, the system optimizes the balance between using enough features for accuracy while limiting computational complexity.
3Ease of manufacture
If a fixed optimal machine learning model is selected for all edge devices, then the deployment simplicity is improved, but the adaptability to different environments and sensor conditions deteriorates
Solution Approach 1:
The system implements dynamic feature selection at inference time, allowing the same base machine learning model to adapt to different environments and sensor conditions by selectively using relevant features. This maintains deployment simplicity of a single model while achieving adaptability through runtime feature selection based on lazy importance measures.
Solution Approach 2:
The system segments the feature set into relevant and irrelevant features for each specific prediction task. By dividing the full feature space and selecting only the necessary segments (features) for each prediction based on computed importance, the system enables a single model to adapt to various conditions without requiring multiple specialized models.
Data Source
AI summary
One example method includes consolidating a data point concerning an autonomous mobile robot that operates in an environment, calculating a lazy feature importance for an available feature of the data point, combining the lazy feature importance with a health score of a sensor that collected data associated with the available feature, to obtain a final feature score for the available feature, selecting a subset of features of the data point, and performing, with a machine learning model, an inference, using only those features in the subset of features, and the inference concerns an aspect of the autonomous mobile robot or the environment.