Sensor Proxy Virtualization for Missing Data Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Power-aware wearable devices face challenges in handling missing sensor data, which can lead to incomplete data sets for AI edge model training and inference, and existing imputation techniques consume excessive power and time.
Innovation Solution
A lightweight sensor proxy virtualization method that uses co-existence probabilities and co-prediction accuracies to train machine learning models, allowing for the prediction and generation of proxy sensor data to replace missing data, thereby enabling uninterrupted multi-sensor edge AI modeling with minimal computational resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If existing imputation techniques are used to handle missing sensor data, then data completeness for AI model training is improved, but power consumption and computation time increase excessively
Solution Approach 1:
The patent creates proxy sensor data copies from available sensor data streams to replace missing data. Instead of using complex imputation techniques, the system generates simplified proxy representations that mimic the statistical properties of missing sensor data, thereby maintaining data completeness while reducing computational overhead and power consumption.
Solution Approach 2:
The system changes the parameters of available sensor data streams by applying transformations and aggregations to create proxy data that statistically represents missing sensor measurements. This involves modifying data parameters such as sampling rates, aggregation windows, and transformation functions to generate equivalent information from different sensor sources.
2Loss of information
If existing imputation techniques are used to handle missing sensor data, then data completeness for AI model training is improved, but computation time increases excessively
Solution Approach 1:
The patent creates proxy sensor data copies from available sensor data streams to replace missing data. Instead of using complex imputation techniques, the system generates simplified proxy representations that mimic the statistical properties of missing sensor data, thereby maintaining data completeness while reducing computational overhead and power consumption.
Solution Approach 2:
The system performs preliminary actions by pre-processing available sensor data to create proxy representations that can be quickly deployed to replace missing data. This includes pre-computing statistical properties, aggregation functions, and transformation parameters from available sensors before data is needed, reducing real-time computation requirements.
3Measurement precision
If sensor data from multiple sensors is used for AI edge model training, then model accuracy is improved, but power dissipation increases due to processing multiple data streams
Solution Approach 1:
The patent makes available sensor data streams multi-functional by using them to both directly train AI models and generate proxy representations for missing sensor data. This universal approach allows the system to leverage data from fewer sensors for multiple purposes, reducing the need to process multiple separate data streams while maintaining model accuracy through the proxy data mechanism.
4Loss of information
If frequent sensor data collection is performed to ensure data availability, then data completeness is improved, but power consumption increases
Solution Approach 1:
The patent introduces proxy sensor data as an intermediary that mediates between available sensor measurements and the need for complete multi-sensor data streams. Instead of frequently activating all sensors, the system uses proxy representations generated from available sensors to fill gaps, reducing the frequency of data collection from multiple sensors while maintaining data availability for AI model training.
Data Source
AI summary
Systems and methods for lightweight proxy virtualization of a plurality of sensor data streams in a device are described. A processor can receive a plurality of sensor data streams from a plurality of sensors. The processor can identify missing sensor data in a sensor data stream among the plurality of sensor data streams. The processor can predict a value of the missing sensor data by running a machine learning model trained using sensor data determined based on at least one of a plurality of co-existence probabilities of the plurality of sensor data streams and a plurality of co-prediction accuracies of the plurality of sensor data streams.