Neural Network Update via Clustering Centroids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image recognition technology requires large amounts of labeled data for machine learning, and the accuracy of these labels directly affects the performance of computer learning models, but existing systems lack an efficient method to determine when the neural network needs updating with new learning materials.
Innovation Solution
An image system and detection method that receives recognition label data sets from terminal devices, calculates matching degree values, sets weight values, classifies label points into cluster groups using a clustering algorithm, and adjusts the neural network based on the centroid of the largest cluster group when specific conditions are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the neural network is continuously updated with new learning materials from multiple terminal devices, then the image recognition accuracy is improved, but the system complexity and computational resources required increase
Solution Approach 1:
The patent segments the learning material collection process by dividing terminal devices into different cluster groups based on clustering algorithms. Each cluster group represents a segmented portion of the overall data source, allowing the system to process and update the neural network with organized, manageable segments rather than handling all terminal devices uniformly, thus improving accuracy while managing system complexity.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and clustering learning materials from terminal devices before updating the neural network. The system collects label images, determines matching degree values, applies clustering algorithms to group terminal devices, and prepares centroid calculations in advance. This preliminary organization of data reduces the computational burden during actual neural network updates, balancing accuracy improvement with system complexity management.
2Reliability
If recognition label data sets from multiple terminal devices are collected and used for updating, then the learning model performance is enhanced, but the time and computational resources required for processing increase
Solution Approach 1:
The patent implements preliminary action by pre-processing learning materials through clustering algorithms before neural network updates. Terminal devices are grouped into clusters based on their label image characteristics, and centroids are calculated in advance. This preliminary organization reduces the processing time during actual updates, as the system only needs to process representative centroid data rather than all individual terminal device data, thus enhancing performance while reducing time loss.
Solution Approach 2:
The patent uses copying by utilizing centroid representations to stand in for entire cluster groups of terminal devices. Instead of processing all original label images from multiple terminal devices, the system creates simplified centroid copies that represent the essential characteristics of each cluster. This copying approach maintains learning model performance while significantly reducing the computational time and resources required for processing.
3Adaptability or versatility
If the system processes and clusters all label images from terminal devices, then the comprehensiveness of learning data is improved, but the computational load and processing complexity increase
Solution Approach 1:
The patent applies segmentation by dividing the comprehensive set of label images from terminal devices into distinct cluster groups using clustering algorithms. Each cluster represents a segmented portion of the overall data distribution, allowing the system to maintain comprehensiveness by covering multiple clusters while reducing computational load by processing only representative samples from each segment rather than all individual images.
Solution Approach 2:
The patent performs preliminary action by pre-clustering terminal devices based on their label image characteristics before processing. The clustering algorithm groups similar devices together, and centroids are calculated in advance to represent each cluster. This preliminary segmentation and representation reduces the computational load during actual learning updates while preserving the comprehensiveness of the learning data through diverse cluster coverage.
Data Source
AI summary
The present disclosure relates to an imaging system and a detection method. The detection method includes the following steps. Receiving, by a processing unit of the image system, multiple recognition label data sets transmitted from multiple terminal devices. Determining a matching degree value between the recognition label data sets and an image data, and obtaining multiple weight values from a storage unit corresponding to the terminal devices. Setting the weight values and the corresponding matching degree values as multiple label points, and classifying the marker points into multiple cluster groups by a clustering algorithm. Calculating a centroid of the largest cluster group. The centroid of the largest cluster group corresponds to a clustering weight value and a clustering matching value. When the clustering weight value or the clustering matching value meets an adjustment condition, adjusting a neural network unit according to the largest cluster group.