Vehicle Learning System with Collaborative Neural Network Accuracy Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vehicle learning systems face challenges in verifying the accuracy of neural network outputs due to insufficient training data, especially in varying situations encountered during vehicle operation, leading to difficulties in determining correct values and maintaining reliable misfire detection in internal combustion engines.
Innovation Solution
A vehicle learning system comprising a first execution device on the vehicle and a second execution device outside the vehicle, which collaboratively acquire and evaluate input data using mapping data to assess the relationship between predetermined variables and output accuracy, allowing for the verification of input variables and potential updates to improve accuracy without complicating the mapping structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning is used to enhance reliability of learned model, then accuracy of output value is improved, but sufficient training data cannot be obtained in various situations
Solution Approach 1:
The patent performs preliminary actions by collecting and storing various types of data (sensor data, operation data, map data) before the neural network is deployed on vehicles. This pre-collected data serves as training data, allowing the model to be trained in advance with diverse situations rather than relying on sufficient training data being available after deployment.
Solution Approach 2:
The patent introduces an intermediary data collection and storage system that acts as a bridge between data sources and the neural network training process. This intermediary system accumulates training data from various sources including sensors, operation logs, and map data, making it available for training without requiring direct access during vehicle operation.
2Measurement precision
If neural network structure is complicated to improve accuracy, then accuracy of output value is improved, but calculation load increases
Solution Approach 1:
The patent segments the neural network into multiple processing stages: a first neural network performs initial processing of input data, and a second neural network performs subsequent processing. This segmentation allows complex accuracy-improving operations to be distributed across multiple simpler networks rather than requiring one large complex network, reducing the calculation load on any single processing unit.
Solution Approach 2:
The patent employs dynamic processing where the neural network adapts its processing based on input characteristics. The system dynamically selects which neural network to use and how to process data based on the situation, allowing the system to maintain high accuracy while reducing calculation load when full neural network processing is not necessary.
Data Source
AI summary
A vehicle learning system includes a first execution device mounted on a vehicle, a second execution device outside the vehicle, and a storage device. The storage device stores mapping data including data, which is learned by machine learning and defines mapping that receives input data based on a detection value of an in-vehicle sensor and outputs an output value. The first execution device and the second execution device execute, in cooperation with each other, an acquisition process of acquiring input data, a calculation process of calculating an output value with the input data as an input of the mapping, and a relationship evaluation process of evaluating a relationship between a predetermined variable different from a variable corresponding to the output value and accuracy of the output value. The first execution device executes at least the acquisition process, and the second execution device executes at least the relationship evaluation process.