IV-GNN : interval valued data handling using graph neural network.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Interval-valued data effectively represents uncertainty but is challenging for standard machine learning models.
• Graph Neural Networks (GNNs) excel with graph data but typically require countable feature spaces.
• A research gap exists in handling interval-valued features within GNNs for non-Euclidean data.

Purpose of the Study:

• To propose a novel Interval-Valued Graph Neural Network (GNN) model.
• To address the limitation of GNNs in processing interval-valued features on graphs.
• To develop a more general GNN applicable to uncountable feature spaces.

Main Methods:

• Introduced a new GNN architecture designed for interval-valued features.
• Developed a novel interval aggregation scheme to capture diverse interval structures.
• Relaxed the restriction of countable feature spaces without increasing time complexity.

Main Results:

• The proposed Interval-Valued GNN demonstrates expressive power in handling interval structures.
• The model achieved competitive performance on graph classification tasks.
• Validated theoretical findings on benchmark and synthetic network datasets.

Conclusions:

• The Interval-Valued GNN successfully bridges the gap between interval analysis and GNNs.
• The model offers a more generalized approach to GNNs, applicable to a wider range of data.
• This work advances the capability of GNNs for complex, real-world datasets with inherent uncertainty.