Graph Data Augmentation via Perturbation for Model Generalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training machine learning models on graph data structures can result in overfitting when the data set is insufficient, leading to poor performance on new data due to the model's alignment with minimal training data.
Innovation Solution
The technique involves augmenting the graph data structure by generating perturbed views through deletion of edges and nodes based on occurrence, importance, and proximity metrics, allowing the model to learn from both original and augmented graphs, thereby reducing overfitting and enhancing generalization and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a machine learning model is trained on a graph data structure with insufficient graphs, then the training process is faster and requires less computational resources, but the model becomes overfitted and performs poorly on new data
Solution Approach 1:
The patent applies preliminary action by generating augmented graphs before model training through systematic perturbations. The augmentation process creates multiple perturbed versions of each graph by deleting nodes and edges based on occurrence, importance, and proximity metrics. This pre-processing step expands the limited training dataset, allowing the model to learn from more diverse examples without requiring additional computational resources during the actual training phase.
Solution Approach 2:
The patent implements parameter changes by systematically modifying graph structure parameters through perturbation. Specifically, nodes and edges are deleted with probabilities determined by their occurrence frequency, importance metrics, and proximity relationships. This transforms the original graph parameters into varied perturbed versions, enabling the model to learn robust representations that generalize better to new data while maintaining training efficiency.
2Reliability
If the graph data structure is augmented by generating more graphs, then the model generalization improves, but the data processing time and computational complexity increase
Solution Approach 1:
The patent applies local quality by selectively perturbing specific parts of the graph rather than uniformly modifying the entire structure. Nodes and edges are deleted based on their local properties such as occurrence frequency, importance metrics, and proximity to other elements. This targeted approach creates diverse augmented graphs while minimizing unnecessary computational operations, thus improving model generalization without proportionally increasing data processing time.
Solution Approach 2:
The patent implements partial action by applying perturbation to only certain portions of each graph based on predefined criteria. Instead of modifying all graphs equally or using excessive perturbation that would create unrealistic variations, the method selectively deletes nodes and edges where it matters most for learning robust representations. This balanced approach achieves sufficient augmentation for improved generalization while controlling the computational overhead.
3Reliability
If nodes and edges are deleted to create perturbed views, then the model learns to handle errors and improves robustness, but the graph structure becomes less complete and may lose important information
Solution Approach 1:
The patent converts the potential harm of information loss into a benefit by deliberately creating perturbed graphs with deleted nodes and edges. These perturbations simulate real-world errors and incomplete data scenarios, teaching the model to handle such situations robustly. The deletion process uses occurrence, importance, and proximity metrics to ensure that while some information is lost, the resulting perturbations are meaningful and beneficial for learning error tolerance.
Solution Approach 2:
The patent applies preliminary action by pre-training the model on perturbed graphs that have had nodes and edges deleted. This exposure to incomplete structures before actual deployment prepares the model to handle real-world imperfections. The systematic deletion based on metrics ensures that the model learns to infer missing information and maintain performance even when critical graph elements are absent, thereby improving robustness without sacrificing essential structural information.
Data Source
AI summary
A computing device may receive a set of user documents. Data may be extracted from the documents to generate a first graph data structure with one or more initial graphs containing key-value pairs. A model may be trained on the first graph data structure to classify the pairs. Until a set of evaluation metrics for the model exceeds a set of deployment thresholds: generating, a set of evaluation metrics may be generated for the model. The set of evaluation metrics may be compared to the set of deployment thresholds. In response to a determination that the set of evaluation metrics are below the set of deployment thresholds: one or more new graphs may be generated from the one or more initial graphs in the first graph data structure to produce a second graph data structure. The first and second graph can be used to train the model.