Tree Model Compression Using Error-Based Node Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number of nodes in a learning tree model with improved prediction accuracy leads to significant data storage challenges, making it difficult to mount the model on IC chips with limited storage capacity, and reducing the data amount of the model can lower prediction accuracy.
Innovation Solution
A data amount compressing method that reads error amounts associated with each node in a learned model and deletes nodes based on these error amounts, prioritizing nodes with higher errors and preventing deletion of nodes with high prediction accuracy, to compress the data while minimizing prediction accuracy loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of nodes in the learning tree model is increased to improve prediction accuracy, then the prediction accuracy is improved, but the data storage requirement increases significantly
Solution Approach 1:
The patent extracts and identifies nodes with high error amounts from the learning tree model, then selectively removes these nodes while preserving nodes with low error amounts. This extraction approach allows the model to maintain prediction accuracy by keeping only the essential nodes that contribute most to accurate predictions, while removing redundant nodes that consume storage space.
Solution Approach 2:
The patent changes the parameter of node selection criteria from a fixed structure-based approach to a dynamic error-amount-based approach. By using error amounts as the selection parameter, the system can adaptively determine which nodes to retain or remove based on their actual contribution to prediction accuracy, rather than following a predetermined structural pattern.
2Quantity of substance
If the data amount of the learned model is reduced to fit IC chip storage capacity, then the storage capacity constraint is satisfied, but the prediction accuracy may be lowered
Solution Approach 1:
The patent implements a feedback mechanism where error amounts are calculated for each node based on their prediction performance. This feedback information is then used to guide the node deletion process, ensuring that nodes critical for prediction accuracy (those with low error amounts) are preserved while nodes with high error amounts are removed. The feedback loop between error calculation and node selection ensures that data reduction does not significantly compromise prediction accuracy.
Solution Approach 2:
The patent transforms the model optimization problem from a structural parameter focus to an error-amount parameter focus. By changing the selection criterion to error amounts, the system can achieve data reduction while maintaining the functional integrity of the model, as the error-amount-based selection directly targets nodes that have the least impact on prediction accuracy.
Data Source
AI summary
A data amount compressing method for compressing a data amount corresponding to a learned model obtained by letting the learning model learn a predetermined data group, the learning model having a tree structure in which multiple nodes associated with respective hierarchically divided state spaces are hierarchically arranged, wherein each node in the learned model is associated with an error amount that is generated in the process of the learning and corresponds to prediction accuracy, and the data amount compressing method includes: a reading step of reading the error amount associated with each node; and a node deleting step of deleting a part of the nodes of the learned model according to the error amount read in the reading step, thereby compressing the data amount corresponding to the learned model.