Multi-stage Neural Network Pattern Recognizer Using Positional Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pattern recognition technologies require extensive research and development to create specific recognizers for each application, making them labor-intensive and costly, and lack the ability to process images in a direct and efficient manner like text processing.
Innovation Solution
An image-based pattern recognizer using a neural network with positional coding, allowing for the creation of generic observers that can recognize various patterns without explicit programming, enabling customization and efficient processing of images, video, and audio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a classic pattern recognition approach with explicit routines and variables is used for each specific recognition task, then the recognition function can be implemented accurately, but the research and development period becomes very long and the cost increases
Solution Approach 1:
The patent applies universality by creating a single generic pattern recognizer architecture that can perform multiple different recognition tasks (OCR, ASR, face recognition, gesture recognition, etc.) through configuration rather than requiring separate specialized recognizers for each task. This universal architecture eliminates the need for lengthy R&D periods for each specific application while maintaining accurate recognition functions.
Solution Approach 2:
The patent uses parameter changes by allowing the pattern recognizer to be customized through modifying parameters such as the observation model, emission model, and transition model rather than changing the fundamental architecture. This enables rapid adaptation to different recognition tasks by adjusting parameters while keeping the core system intact, significantly reducing development time.
2Reliability
If a classic pattern recognition approach with explicit routines and variables is used for each specific recognition task, then the recognition function can be implemented accurately, but the cost of resulting applications increases
Solution Approach 1:
The universal pattern recognizer architecture allows a single system to handle multiple recognition tasks, reducing the need for multiple separate development projects and reducing overall development costs while maintaining accurate recognition performance across different applications.
Solution Approach 2:
The patent employs copying by using the same underlying architecture and algorithms for different recognition tasks, copying the proven successful approach across multiple applications rather than developing new solutions from scratch for each task, thereby reducing costs.
3Productivity
If image-based pattern recognition is performed without positional coding, then the processing cannot be done in a direct manner like text processing, but the relationship between raw data and search patterns remains indirect
Solution Approach 1:
The patent introduces positional coding as an intermediary representation that bridges the gap between raw image data and pattern recognition. By encoding spatial positions explicitly, the system creates a direct relationship between image data and search patterns, enabling efficient processing similar to text while preserving spatial information.
Data Source
AI summary
An image-based pattern recognizer and a method and apparatus for making such a pattern recognizer are disclosed. By employing positional coding, the meaning of any feature present in an image can be defined implicitly in space. The pattern recognizer can be a neural network including a plurality of stages of observers. The observers are configured to cooperate to identify the presence of features in the input image and to recognize a pattern in the input image based on the features. Each of the observers includes a plurality of neurons. The input image includes a plurality of units, and each of the observers is configured to generate a separate output set that includes zero or more coordinates of such units.