Model Parameter Learning with Alternating Noise Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model parameter learning devices, such as those used in autonomous driving vehicles, are susceptible to noise in image data, leading to unstable performance when environmental conditions change, as they rely solely on error backpropagation with mean square error as a loss function.
Innovation Solution
A model parameter learning device that creates noise-added data and uses teacher data to minimize errors between model outputs and teacher data, incorporating multiple learning processes with alternating noise inputs to suppress noise influence, including the use of separate models and noise types to optimize parameter learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If model parameters are learned by error backpropagation with mean square error as loss function, then learning process is simple, but output data becomes susceptible to noise in input data
Solution Approach 1:
The learning process is segmented into two distinct phases: first learning processing that minimizes error between second model data and teacher data, and second learning processing that minimizes loss function containing third model data. This segmentation allows each phase to focus on different aspects of noise suppression while maintaining overall system simplicity
Solution Approach 2:
Teacher data acts as an intermediary between noisy input data and model parameters. By introducing teacher data that corresponds to characteristic data, the system can learn to suppress noise influence without requiring complex learning algorithms, as the teacher data provides a reliable reference target
2Adaptability or versatility
If noise components in image data increase due to environmental changes, then model adaptability to real-world conditions improves, but travel state control stability deteriorates
Solution Approach 1:
The system performs preliminary noise suppression learning by minimizing error between second model data and teacher data before executing control tasks. This preliminary action embeds noise resistance into model parameters in advance, ensuring stable travel state control even when environmental conditions change and noise increases
Solution Approach 2:
The system changes learning parameters by using different loss functions for different learning phases. First learning uses error minimization with teacher data, while second learning uses loss function minimization with third model data. This parameter change enables the model to adapt to various noise levels while maintaining control stability
Data Source
AI summary
Provided is a model parameter learning device and the like capable of learning model parameters such that the influence of a noise in input data can be suppressed. A model parameter learning device (1) alternately carries out first learning processing for learning model parameters W1, b1, W2 and b2 such that an error between data Xout and data Xorg is minimized, and second learning processing for learning model parameters W1, b1, Wm, bm, Wq and bq such that a loss function LAE is minimized.