Neural Network Intermediate Layer Adjustment via Side Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multilayer neural networks face challenges in achieving 100% classification accuracy due to variability in learning data quality and quantity, leading to inconsistent performance in identification and classification tasks.
Innovation Solution
A classification device and method that incorporates a side information calculating unit and a multilayer neural network, where side information is used to adjust the output values of the intermediate layer, allowing for reprocessing when discrepancies occur between processing results and side information, thereby improving classification accuracy without requiring relearning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If learning data quality and quantity are increased to improve classification accuracy, then classification accuracy rate is improved, but learning cost and time are increased
Solution Approach 1:
The patent applies preliminary action by calculating side information (such as image quality metrics, sensor conditions, or contextual data) before performing classification. This pre-calculated information is then used to adjust the neural network's output or guide post-processing, allowing the system to achieve higher accuracy without requiring additional learning time or data collection.
Solution Approach 2:
The patent introduces side information as an intermediary element that mediates between the raw classification output and the final result. This intermediary information (such as confidence scores, quality metrics, or contextual parameters) allows the system to refine classifications without retraining the neural network, thereby improving accuracy without increasing learning time.
2Measurement precision
If learning data quality and quantity are increased to improve classification accuracy, then classification accuracy rate is improved, but cost is increased
Solution Approach 1:
The patent calculates side information in advance that can be used to enhance classification accuracy without requiring additional learning data. By pre-processing or pre-calculating auxiliary information (such as image characteristics, sensor metadata, or environmental context), the system achieves better performance without increasing the quantity of training data needed.
Solution Approach 2:
The side information acts as an intermediary that bridges the gap between limited training data and high classification accuracy. Instead of relying on more training data, the system uses this intermediary information to refine and adjust classification results, achieving improved accuracy with the same or reduced learning data requirements.
3Measurement precision
If relearning is performed on the multilayer neural network to improve classification accuracy, then classification accuracy rate is improved, but processing time is increased
Solution Approach 1:
The patent implements feedback by comparing the neural network's classification output with pre-calculated side information and using this comparison to adjust the final classification result. This feedback mechanism allows the system to improve accuracy on a per-classification basis without requiring time-consuming relearning or retraining of the entire neural network.
Solution Approach 2:
By calculating side information in advance and using it to guide or adjust classification results in real-time, the system avoids the need for time-consuming relearning. The preliminary calculation of auxiliary information enables rapid accuracy improvement without extending processing time for retraining the model.
Data Source
Figure 1~2
Figure 3A~4
Figure 5
AI summary
A side information calculating unit (110) calculates side information for assisting either identification processing or classification processing. When there is a discrepancy between a processing result of either the identification processing or the classification processing, and the side information, the multilayer neural network (120) changes an output value of an intermediate layer (20) and performs either the identification processing or the classification processing again.