Pruned Predicate List for Logical Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training logical neural networks (LNNs) with a large number of predicates leads to an exponential increase in training time, causing processing strain and inefficiency, as conventional approaches lack techniques for predicate selection and control.
Innovation Solution
The method involves extracting predicates from sentences, using an explainer component to determine attentions, and a pruner model to generate a pruned list of predicates, which are then used to train the LNN, reducing the number of predicates and thereby decreasing training time while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of predicates are used to train the LNN, then the model accuracy is improved, but the training time increases exponentially
Solution Approach 1:
The patent extracts and selects only the most important predicates from the complete set of extracted predicates. An explainer component analyzes sentences to determine attentions from predicates, and a pruner model uses these attentions to generate a pruned list of predicates. This extraction process removes redundant or less important predicates while retaining the most impactful ones for training, thereby reducing training time while maintaining model accuracy.
Solution Approach 2:
The patent applies local quality by differentiating the importance of different predicates. Instead of treating all predicates equally, the system assigns different weights or selections based on their individual impact on the training task. The explainer component and pruner model identify and prioritize predicates with higher attentions, creating a customized subset that optimizes training efficiency for each specific task.
2Loss of information
If all extracted predicates are used for training, then the model comprehends more information, but the processing resources are overutilized
Solution Approach 1:
The system extracts and removes redundant predicates from the complete set, retaining only those that provide the most valuable information for training. The pruner model, guided by attentions from the explainer component, identifies and eliminates predicates that do not contribute significantly to the training objective, thereby reducing processing resource consumption while preserving essential information.
Solution Approach 2:
The patent changes the parameter of predicate selection from using all extracted predicates to using a curated subset. By modifying the selection criteria through the explainer and pruner components, the system optimizes the balance between information completeness and processing efficiency, adjusting the predicate list to match the actual training needs.
Data Source
AI summary
A computer-implemented method, according to one embodiment, includes extracting predicates from a predetermined plurality of sentences, and causing an explainer component to analyze the sentences to determine attentions from the predicates of the sentences. The method further includes causing the extracted predicates to be input into a predetermined pruner model. The pruner model is trained to use the attentions to generate a pruned list of predicates from the extracted predicates. A logical neural network is caused to be trained using the pruned list of predicates. A computer program product, according to another embodiment, includes a computer readable storage medium having program instructions embodied therewith. The program instructions are readable and/or executable by a computer to cause the computer to perform the foregoing method.