Tree-Structured Model Integration for Comprehensive Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning techniques struggle to integrate learned models effectively due to their nature, hindering the generation of a comprehensive prediction model.
Innovation Solution
A method and system for integrating learned models with tree structures by duplicating nodes where necessary, associating learning-related parameters, and performing weighted addition based on data quantities to generate a single, comprehensive model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple learned models are integrated using conventional machine learning techniques, then comprehensive prediction capability is improved, but integration difficulty increases due to the nature of neural networks
Solution Approach 1:
The learned model is segmented into discrete tree nodes representing different state spaces. Each node can be independently integrated with other nodes from different models, enabling modular integration without requiring integration of entire complex neural network structures. This segmentation transforms the integration problem from a global complex task into localized node-level operations.
Solution Approach 2:
When a node exists only in one learned model, it is duplicated into the integrated model structure. This copying mechanism allows the integrated model to inherit all necessary nodes and their associated data from individual models, preserving prediction capabilities while establishing a unified model structure that combines multiple sources.
2Reliability
If nodes are duplicated during integration when they exist in only one model, then integration completeness is improved, but processing efficiency improves by avoiding redundant operations
Solution Approach 1:
Before performing integration operations, the system preliminarily determines which nodes exist in each model and identifies nodes requiring duplication. This preliminary analysis enables the integration process to directly duplicate only necessary nodes without attempting to process or compare non-existent nodes, significantly improving processing efficiency while ensuring all required nodes are included in the integrated model.
3Measurement precision
If learning-related parameters are integrated using weighted addition based on data quantities, then prediction accuracy is improved, but calculation complexity increases
Solution Approach 1:
The system changes the integration parameter from simple uniform addition to weighted addition based on data quantities. Each node's learning-related parameters (such as error amounts) are integrated with weights proportional to the number of data points in the corresponding state space. This parameter change enables the integrated model to prioritize nodes with more representative data, improving prediction accuracy while maintaining a relatively simple weighted averaging calculation framework.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
[Problem to be Solved] Multiple learned models obtained using a new machine learning framework having a tree structure are integrated to generate a single learned model. [Solution] Provided is a learned model integration method for integrating multiple different learned models obtained by letting a learning model learn a predetermined data group, the learning model having a tree structure, the method comprising: a data reading step of reading data related to the multiple different learned models from a predetermined memory unit; and an integrating step in which, for each node constituting a tree structure related to the multiple different learned models, when a node exists in only one learned model, the node is duplicated, and when nodes exist in corresponding positions in the multiple learned models, the corresponding nodes are integrated, thereby integrating the multiple different learned models into a single learned model.