Machine Learning Classifier Guide Map Update Mechanism
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning classifiers can produce suboptimal results when faced with specific patterns in images, leading to incorrect classifications, necessitating a method to prevent wrong learning and support learning for difficult images.
Innovation Solution
A machine learning device and method that uses a guide map to update the classifier, comparing intermediate data with the guide map and first class information to refine the classification process, ensuring accurate classification even with images containing hindering patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning classifier performs iterative learning on images with specific patterns, then classification performance deteriorates due to wrong learning, but if learning is avoided for these images, then productivity decreases
Solution Approach 1:
The patent introduces a guide map as an intermediary element that guides the machine learning classifier during the learning process. The guide map highlights relevant regions in images, enabling the classifier to focus on important features while ignoring harmful patterns. This mediator allows the system to maintain high classification accuracy without sacrificing learning productivity, as the guide map provides directional guidance rather than blocking entire images from learning.
2Measurement precision
If machine learning classifier processes all images uniformly, then device complexity is reduced, but measurement precision deteriorates due to inability to distinguish relevant from irrelevant features
Solution Approach 1:
The patent applies local quality by generating guide maps that provide different weighting or attention levels to different regions of images. Instead of treating all image regions uniformly, the guide map highlights specific local areas that are relevant for classification while downplaying or ignoring regions with harmful patterns. This allows the classifier to achieve high measurement precision by focusing computational resources on discriminative features without requiring complex preprocessing or multiple processing paths.
3Reliability
If machine learning classifier uses traditional updating methods, then ease of operation is maintained, but reliability deteriorates due to wrong learning from harmful patterns
Solution Approach 1:
The patent implements preliminary action by generating guide maps before the machine learning classifier processes images. The guide maps are created in advance to identify and highlight relevant regions while marking harmful patterns for exclusion. This preliminary guidance is integrated into the standard learning workflow, allowing the classifier to operate with enhanced robustness without requiring complex operational changes or manual intervention during the learning process.
Data Source
AI summary
A machine learning method of a machine learning device including a machine learning classifier includes receiving, at the machine learning device, an image and first class information associated with the image, generating, at the machine learning device, second class information associated with the image by performing classification on the image by using the machine learning classifier, and as the second class information is generated, updating, at the machine learning device, the machine learning classifier by performing a first learning operation when a guide map is received together with the image and performing, at the machine learning device, a second learning operation different from the first learning operation when the guide map is not received together with the image.