Coupled Artificial Learning Units for Fast Classification and Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence systems, particularly neural networks, are limited to specialized tasks and require extensive retraining for different applications, leading to inefficiencies and impracticality in real-time responses due to high-dimensional spaces and complexity.
Innovation Solution
A method involving two coupled artificial learning units, where one unit influences the parameters of another through modulation functions, allowing for rapid classification and in-depth analysis without direct feedback, utilizing asymmetric classification memories and potentially time-dependent superpositions to enhance efficiency and responsiveness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single neural network is used for specialized tasks, then the network can be trained effectively for that specific application, but the network becomes useless for other areas and requires complete retraining or different approaches for different applications
Solution Approach 1:
The patent divides a single large neural network into multiple specialized sub-networks, each trained for specific tasks. These sub-networks can be selectively activated based on the input task, enabling the system to handle multiple applications without complete retraining. The segmentation allows each sub-network to specialize while the overall system maintains versatility.
Solution Approach 2:
The patent creates a universal neural network system that can perform multiple functions through a collection of specialized sub-networks. The system includes a task classification component that routes inputs to appropriate sub-networks, and a mechanism to combine their outputs, achieving multi-functionality without requiring each individual network to be universally applicable.
2Adaptability or versatility
If the dimensionality and complexity of AI systems are increased to achieve universal applicability, then the system can handle more applications, but real-time responses become impossible due to exponentially increasing training and test datasets
Solution Approach 1:
The patent segments the processing task across multiple smaller sub-networks rather than using one large complex network. Each sub-network processes specific aspects of the input independently and in parallel, significantly reducing the computational complexity and data volume each unit must handle, thereby enabling real-time responses while maintaining universal applicability.
Solution Approach 2:
The patent implements a task classification component that selectively activates only the relevant sub-networks needed for a given input, rather than always engaging the entire system. This partial action approach reduces the effective computational load for each specific task, enabling faster response times while maintaining the capability for universal applicability when needed.
3Measurement precision
If extensive retraining is performed for different applications, then the system can be optimized for each specific task, but the process becomes inefficient and impractical
Solution Approach 1:
The patent segments the training process into independent sub-networks, each trained once for its specific function. This eliminates the need for extensive retraining when switching applications, as the appropriate pre-trained sub-networks are simply selected and activated. The segmentation maintains task optimization while dramatically improving training efficiency.
Solution Approach 2:
The patent performs the training action in advance for each specialized sub-network during a setup phase. Once trained, these sub-networks are ready for immediate deployment without requiring retraining for different applications. The preliminary training action enables both task optimization and high productivity during operational use.
Data Source
Figure 1
Figure 2a~2h
Figure 3
AI summary
The invention relates to a method in a system formed by at least two artificial learning units, comprising: inputting input values at at least one first artificial learning unit and a second artificial learning unit; obtaining first output values from the first artificial learning unit; forming one or more modulation functions on the basis of the output values of the first artificial learning unit; applying the formed one or more modulation functions to one or more parameters of the second artificial learning unit, wherein the one or more parameters influence the processing of input values and the obtaining of output values in the second artificial learning unit; and finally, obtaining second output values from the second artificial learning unit.