Adversarial Sensor Encoding for Redundant IoT Data Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for machine learning with IoT devices face challenges in efficiently reducing data dimensions and eliminating redundancy in sensor data from multiple instruments, leading to decreased communication efficiency due to redundant information.
Innovation Solution
A machine learning device that includes an acquisition unit, encoding units, an estimation unit, and adversarial estimation units, trained through machine learning to encode and eliminate redundancy in sensor data from multiple measuring devices, using models like deep neural networks and adversarial estimation models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If sensor data from multiple measuring devices is processed independently through separate neural networks, then each device can perform local processing, but redundant information is introduced in the low-dimensional observation data
Solution Approach 1:
The patent merges the independent neural networks of multiple measuring devices into a single shared neural network that processes sensor data from all devices. This unified approach eliminates redundant information by ensuring consistent feature extraction across all measuring devices while maintaining the ability to process data locally at each device before transmission to the central processing unit.
Solution Approach 2:
The shared neural network serves as a universal processing model that handles data from multiple different measuring devices. This universal model can process various types of sensor data (acceleration, angular velocity, temperature, humidity) and transform them into a common representation space, eliminating device-specific redundancies while preserving essential information.
2Loss of information
If high-frequency and large-capacity data is transmitted from IoT devices, then more comprehensive information is available, but communication power consumption increases and communication capacity is exceeded
Solution Approach 1:
The patent performs preliminary dimensionality reduction and feature extraction at the sensor data generation stage using the shared neural network. By pre-processing the data locally and transforming it into a compact representation with essential features before transmission, the system minimizes the amount of data that needs to be transmitted over the communication channel, thereby reducing communication power consumption while preserving information completeness.
Solution Approach 2:
The neural network transforms high-dimensional sensor data into a lower-dimensional observation space by learning optimal feature representations. This dimensional transformation compresses the data while retaining the essential information needed for accurate predictions, enabling efficient communication with reduced data volume and associated energy consumption.
3Quantity of substance
If the number of nodes in the intermediate layer is reduced for dimensionality reduction, then communication data amount decreases, but information loss may occur
Solution Approach 1:
The patent optimizes the number of nodes in the intermediate layer of the neural network by training the model to achieve accurate predictions. Through parameter learning during training, the system determines the optimal dimensionality reduction level that balances data compression with information preservation, ensuring that the reduced-dimensional observation data contains sufficient information for accurate task performance.
Data Source
AI summary
A machine learning device that trains a first encoding model for encoding first sensor data into first code, a second encoding model for encoding second sensor data into second code, and an estimation model for making estimation using the first code and the second code such that an estimation result from the estimation model conforms to correct answer data, trains a first adversarial estimation model that outputs an estimated value of the second code in response to the input of the first code such that the estimated value of the second code estimated by the first adversarial estimation model conforms to the second code outputted from the second encoding model, and trains the first encoding model such that the estimated value of the second code estimated by the first adversarial estimation model does not conform to the second code outputted from the second encoding model.