SparsePool: A Graph Pooling Framework via Sparse Representation for Graph Classification.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Graph neural networks (GNNs) excel at graph classification, often using graph pooling.
• Current pooling methods struggle with global structural information, losing substructures and interpretability.
• Limitations are significant in molecular analysis and social network mining.

Purpose of the Study:

• Introduce SparsePool, a graph pooling method addressing limitations of existing approaches.
• Integrate node features and structural patterns via atomic decomposition for enhanced graph representation.
• Improve interpretability and preserve semantically meaningful substructures in graph data.

Main Methods:

• Propose SparsePool, a graph pooling method utilizing atomic decomposition.
• Employ Boolean matrix factorization to dynamically decompose graphs into interpretable atomic units.
• Introduce an Atomic Pooling Neural Network (APNN) for graph representation learning.

Main Results:

• SparsePool outperforms state-of-the-art pooling methods on benchmark datasets.
• Achieved an average classification accuracy improvement of 1.03% over baseline models.
• Demonstrated reduced structural information loss compared to existing methods.

Conclusions:

• SparsePool effectively preserves semantically meaningful substructures and enhances interpretability.
• The method offers a promising direction for graph representation learning in molecular and social network analysis.
• Compatibility with quantum computing paradigms suggests future scalability for industrial applications.