Autoencoder-Based Neural Network Training with Reduced Supervised Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stably training a neural network without a large amount of supervised data is challenging, as existing methods require extensive data for network optimization, especially when adapting to new objects or three-dimensional images.
Innovation Solution
The network training device employs autoencoders to independently train each network layer, reducing the need for supervised data by using lower-dimensional signals for training, allowing for stable training with minimal data through autoencoder mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large amount of supervised data is used for training, then network training stability and accuracy are improved, but data collection cost and time consumption increase
Solution Approach 1:
The patent divides the neural network into multiple independent network layers and trains each layer separately using autoencoders. This segmentation allows each layer to be trained independently with reduced data requirements, thereby maintaining training stability without needing a large amount of supervised data for the entire network.
Solution Approach 2:
The patent introduces autoencoders that transform input data into lower-dimensional latent representations. By changing the dimensionality of the data representation, the system can capture essential features with fewer samples, reducing the quantity of supervised data needed while maintaining training reliability.
2Manufacturing precision
If traditional network training methods are used, then training accuracy can be achieved, but the process requires extensive supervised data which increases system complexity
Solution Approach 1:
The patent employs autoencoders that can train network layers without requiring external supervised data. The autoencoders learn representations from unsupervised data, enabling the system to train itself with minimal external data intervention, thereby reducing data collection requirements while maintaining training accuracy.
Solution Approach 2:
The patent introduces autoencoders as intermediary components between the input data and the neural network layers. These autoencoders serve as mediators that transform and compress data into useful representations, allowing the network to achieve accurate training without directly requiring extensive supervised data.
3Quantity of substance
If network layers are trained independently using autoencoders, then supervised data requirements are reduced, but training time for each layer increases
Solution Approach 1:
By segmenting the network into independent layers that can be trained separately, the patent enables parallel processing of training tasks. Multiple network layers can be trained simultaneously using the same autoencoder framework, reducing the total training time despite the increased complexity of independent training procedures.
Data Source
AI summary
According to an embodiment, a network training device includes a first training unit that trains a first network that converts an input signal to a first signal, a second training unit that trains a second network that converts the first signal to a second signal, and a third training unit that trains a third network that converts the second signal to an output signal. The first training unit trains the first network as an encoder of a first autoencoder. The second training unit trains the second network by backpropagation by using a second signal for training corresponding to the first signal for training as supervised data. The second signal for training is generated by an encoder of a second autoencoder that encodes a third signal for training into the second signal for training, and decodes the second signal for training into the third signal for training.