Neural Network Input Adaptation Using Sensor Response Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks face reduced accuracy and increased processing time when input data from sensors differs from the data sets used during training, necessitating inefficient pre-processing that increases power consumption.
Innovation Solution
A neural network device that includes a sensor response network adaptively trained to the sensor, converting raw data into feature maps suitable for a target neural network, bypassing the need for separate pre-processing units and directly inputting these feature maps into hidden layers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If raw data from sensors is directly input to the neural network, then processing speed increases, but accuracy decreases due to data format mismatch
Solution Approach 1:
The patent applies preliminary action by pre-processing raw sensor data through a dedicated preprocessing network before it enters the main neural network. This preprocessing step converts raw data into appropriately formatted input data, ensuring that the main neural network receives data in the correct format without requiring additional training, thus maintaining both speed and accuracy
2Measurement precision
If pre-processing units are added to convert sensor data, then accuracy improves, but device complexity and power consumption increase
Solution Approach 1:
The patent merges the preprocessing function into an integrated preprocessing network that is coupled with the main neural network. This unified architecture performs both preprocessing and neural network operations in a coordinated manner, reducing overall system complexity compared to separate preprocessing units while maintaining high processing accuracy
3Measurement precision
If pre-processing units are added to convert sensor data, then accuracy improves, but power consumption increases
Solution Approach 1:
The patent implements self-service by designing the preprocessing network to automatically adapt to different sensor types and data formats without requiring external training or configuration. The preprocessing network autonomously learns the appropriate transformations through reinforcement learning during operation, eliminating the need for additional training processes that would consume extra power while maintaining high accuracy
Data Source
AI summary
The inventive concept relates to a neural network device, an operation method of the neural network device, and a neural network system including the neural network device, the operation method of the neural network device includes, receiving raw data from a sensor, converting the raw data to a feature map suitable for a first hidden layer among a plurality of hidden layers of a target neural network using a sensor response network adaptively trained to the sensor, inputting the feature map to the first hidden layer of the target neural network, and driving the neural network on a basis of the feature map.