Neural Network Training Using a Parallel Teacher Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training neural networks in safety-critical systems are slow due to complex test runs and require large data transmission, necessitating a faster development cycle and reduced data transfer.
Innovation Solution
A method involving a first neural network trained on sensor signals using a second neural network with a different architecture, allowing parallel operation and adjustment based on output comparisons, which can be performed on embedded computing architectures with reduced computing and memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex test runs and validation runs are conducted to train neural networks in safety-critical systems, then the reliability and safety of the system is improved, but the development cycle becomes slow and data transmission requirements increase
Solution Approach 1:
The patent segments the neural network training process into two distinct networks: a first neural network for safety-critical decision-making and a second neural network for generating training data. This segmentation allows parallel processing where the second network generates training examples without interfering with the operational first network, thereby accelerating development while maintaining safety standards
Solution Approach 2:
The second neural network performs preliminary actions by generating training data and validation examples in advance. This preliminary data generation enables the first neural network to be trained more efficiently without requiring extensive real-world test runs, thus reducing the development cycle while maintaining reliability
2Measurement precision
If complex test runs are conducted to validate neural network behavior, then the measurement precision of system behavior is improved, but the quantity of data that must be transmitted increases
Solution Approach 1:
The patent extracts the training data generation function from the operational neural network by implementing a separate second neural network. This extraction allows the system to generate training data locally without transmitting large volumes of operational data to external data centers, reducing data transmission requirements while maintaining behavior analysis accuracy
Solution Approach 2:
The second neural network acts as an intermediary that generates synthetic training data and validation examples. This intermediary enables precise measurement of neural network behavior through controlled test cases without requiring transmission of massive datasets, thus reducing data transmission volume while maintaining measurement precision
Data Source
AI summary
A method for training a first neural network is disclosed. The neural network is configured to ascertain, based on sensor signals of a technical system, an output signal characterizing a classification and/or a regression result regarding the sensor signal. The method includes (i) during operation of the technical system, receiving a sensor signal of the technical system, (ii) ascertaining a first output signal by way of the first neural network and based on the sensor signal, (iii) ascertaining a second output signal by way of a second neural network and based on the sensor signal, wherein the second neural network has a different architecture than the first neural network, and (iv) training the first neural network by adjusting parameters of the first neural network, wherein the first neural network is trained as a function of the second output signal.