Generative Graph Modeling for Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Datasets with missing cell values lead to poor predictive performance, bias, and incorrect analyses in machine learning models, as existing data augmentation methods often exclude incomplete data points, resulting in biased and non-representative visualizations and predictions.
Innovation Solution
A data augmentation system using a graph model that computes probabilities for additional edges based on nonnegative matrix factorization, representing both homophilous and heterophilous clusters, to predict and fill missing values in datasets, thereby generating an augmented dataset that includes all data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data points with missing cell values are excluded from the dataset, then the data visualization model can generate accurate visualizations, but the analysis becomes biased and loses generalizability
Solution Approach 1:
The system performs preliminary data augmentation by predicting and filling missing cell values before the data visualization model processes the dataset. This preliminary completion of incomplete data points allows the model to utilize all available data while maintaining accuracy, thereby resolving the contradiction between excluding incomplete data for accuracy versus including it for generalizability.
2Adaptability or versatility
If data points with missing cell values are included in the dataset, then the analysis improves generalizability, but the visualizations become inaccurate
Solution Approach 1:
The system introduces an intermediary data augmentation component that processes incomplete data points by predicting missing values. This intermediary layer acts as a mediator between the raw incomplete dataset and the visualization model, transforming incomplete data into complete data while preserving the original data's informational content, thus enabling both generalizability and accuracy.
3Adaptability or versatility
If existing data augmentation methods are used to fill missing values, then incomplete data points can be included, but the predictive performance remains poor due to bias
Solution Approach 1:
The system changes the parameters of the data augmentation approach by using a graph-based generative model with homophilous and heterophilous cluster representations instead of traditional imputation methods. This parameter change in the augmentation strategy enables the system to capture complex data relationships and generate more accurate predictions, thereby improving reliability while maintaining the ability to include incomplete data points.
Data Source
AI summary
Systems and methods for data augmentation are described. Embodiments of the present disclosure receive a dataset that includes a plurality of nodes and a plurality of edges, wherein each of the plurality of edges connects two of the plurality of nodes; compute a first nonnegative matrix representing a homophilous cluster affinity; compute a second nonnegative matrix representing a heterophilous cluster affinity; compute a probability of an additional edge based on the dataset using a machine learning model that represents a homophilous cluster and a heterophilous cluster based on the first nonnegative matrix and the second nonnegative matrix; and generate an augmented dataset including the plurality of nodes, the plurality of edges, and the additional edge.