Neural Network Auxiliary Input Layer Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks face challenges in processing operation data with high accuracy when only part of the data obtained during learning processes is available, as they are limited by the data types accessible in the operation mode.
Innovation Solution
A neural network system with a primary input layer and an auxiliary input layer, along with first and second partial networks, that implement pretraining and percolative learning to produce the same calculation results using either both or solely the primary input layer's output, adjusting the influence of auxiliary data through a switcher's non-permeability coefficient.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network uses both primary and auxiliary data during learning, then learning accuracy is improved, but operation accuracy deteriorates when auxiliary data is not available
Solution Approach 1:
The patent segments the neural network into two distinct parts: a first partial network that processes auxiliary data and a second partial network that processes primary data. This segmentation allows the system to learn using both data types while enabling operation with only primary data, resolving the contradiction between learning accuracy and operation reliability.
Solution Approach 2:
The patent extracts the auxiliary data processing capability into a separate first partial network that can be temporarily activated during learning. This extraction allows the system to benefit from auxiliary data during training while being able to operate solely on primary data when needed, maintaining both learning and operation accuracy.
2Adaptability or versatility
If a neural network is designed to operate with limited data types, then operation versatility is improved, but learning capability deteriorates
Solution Approach 1:
The patent creates a universal neural network system that can perform multiple functions: learning with both primary and auxiliary data, and operation with only primary data. The switcher mechanism enables the system to adapt its configuration based on the available data, achieving both learning capability and operation versatility.
Solution Approach 2:
The patent introduces dynamic switching between different network configurations through the switcher. During learning mode, the switcher activates both partial networks to utilize auxiliary data. During operation mode, the switcher dynamically adjusts to use only the second partial network with primary data, making the system adaptable to different operational requirements.
3Productivity
If auxiliary data is used during learning, then learning speed is improved, but system complexity increases
Solution Approach 1:
The patent segments the neural network into two partial networks with distinct functions. The first partial network handles auxiliary data processing while the second handles primary data processing. This segmentation allows efficient parallel processing during learning (improving learning speed) while maintaining a modular structure that manages system complexity through clear functional separation.
Data Source
AI summary
A neural network system includes a primary input layer configured to acquire data in both a learning mode and an operation mode, an auxiliary input layer configured to acquire data solely in the learning mode, a first partial network configured to carry out learning using both an output of the primary input layer and an output of the auxiliary input layer and to subsequently carry out learning solely using the output of the primary input layer so as to produce a same calculation result as a calculation result produced using both the output of the primary input layer and the output of the auxiliary input layer, and a second partial network configured to carry out calculations in the learning mode and the operation mode upon receiving an output of the first partial network.