Sensor Node Active Learning for Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing environmental sensing systems face challenges in minimizing data communication and labeling costs while maintaining identification accuracy, particularly due to changes in environments, sensor node deterioration, and positional deviations, which require frequent updates to identification models.
Innovation Solution
A sensor node and server device system that employs active learning to select and transmit sensor data with low certainty for labeling, using a certainty degree calculation unit to identify uncertain data and a transmission unit to send this data to the server for updating the identification model, thereby reducing communication and manpower costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all sensor data is transmitted to the server for labeling, then the identification model can be updated comprehensively, but the communication cost and power consumption increase significantly
Solution Approach 1:
The patent applies partial action by transmitting only a subset of sensor data (specifically, data with low certainty degrees) to the server for labeling, rather than transmitting all sensor data. This selective transmission reduces communication cost and power consumption while still maintaining identification model accuracy by focusing on the most valuable data for model improvement.
Solution Approach 2:
The patent applies local quality by differentiating sensor data based on their certainty degrees and treating them differently. Data with low certainty degrees are selected for transmission and labeling, while data with high certainty degrees are processed locally without transmission. This differentiated treatment optimizes the balance between model update quality and resource consumption.
2Measurement precision
If manual labeling is performed on all sensor data, then the identification model achieves high accuracy, but the manpower cost increases significantly
Solution Approach 1:
The patent applies partial action by performing manual labeling only on a subset of sensor data that has low certainty degrees, rather than labeling all sensor data. This selective labeling approach maintains identification accuracy by focusing on data that will have the greatest impact on model improvement while significantly reducing the manpower cost associated with labeling.
Solution Approach 2:
The patent applies self-service by using the certainty degree calculation unit to automatically identify and select which sensor data should be transmitted for labeling. This automated selection process reduces the need for manual judgment in data selection, allowing the system to efficiently determine which data requires human labeling attention.
3Adaptability or versatility
If sensor data is transmitted frequently for model updates, then the identification model adapts to environmental changes, but the communication bandwidth consumption increases
Solution Approach 1:
The patent applies partial action by transmitting only the necessary subset of sensor data (data with low certainty degrees) for model updates, rather than transmitting all sensor data frequently. This selective transmission maintains model adaptability to environmental changes while significantly reducing communication bandwidth consumption.
Solution Approach 2:
The patent applies parameter changes by using the certainty degree as a selection criterion to determine which sensor data should be transmitted. By changing the transmission decision from a fixed frequency or volume-based approach to a quality-based approach (using certainty degree as the parameter), the system optimizes the balance between model adaptability and communication bandwidth usage.
Data Source
AI summary
A sensor node connected to a server device through a communication network includes an acquisition unit configured to acquire sensor data from a sensor included in the sensor node, an identification unit configured to classify the sensor data into a plurality of classes to identify occurrence of a predetermined event based on the sensor data and an identification model created by the server device in advance, a certainty degree calculation unit configured to calculate a degree of certainty of the sensor data as an indicator representing uncertainty of a result of the identification based on the result of the identification, and a transmission unit configured to select sensor data having a minimum degree of certainty from a transmission buffer in which the sensor data and the degree of certainty of the sensor data are stored in association with each other, and transmit the selected sensor data to the server device.