Discriminative structural graph classification
arXiv:1905.13422
Abstract
This paper focuses on the discrimination capacity of aggregation functions: these are the permutation invariant functions used by graph neural networks to combine the features of nodes. Realizing that the most powerful aggregation functions suffer from a dimensionality curse, we consider a restricted setting. In particular, we show that the standard sum and a novel histogram-based function have the capacity to discriminate between any fixed number of inputs chosen by an adversary. Based on our insights, we design a graph neural network aiming, not to maximize discrimination capacity, but to learn discriminative graph representations that generalize well. Our empirical evaluation provides evidence that our choices can yield benefits to the problem of structural graph classification.
References in corpus (15)
- Semi-Supervised Classification with Graph Convolutional Networks
- Inductive Representation Learning on Large Graphs
- Neural Message Passing for Quantum Chemistry
- Spectral Networks and Locally Connected Networks on Graphs
- Molecular Graph Convolutions: Moving Beyond Fingerprints
- Fast Graph Representation Learning with PyTorch Geometric
- Biological network comparison using graphlet degree distribution
- Graph Neural Networks with convolutional ARMA filters
- Learning Structural Node Embeddings Via Diffusion Wavelets
- How Powerful are Graph Neural Networks?
- Learning Deep Embeddings with Histogram Loss
- Order Matters: Sequence to sequence for sets
- Universal Invariant and Equivariant Graph Neural Networks
- GraKeL: A Graph Kernel Library in Python
- Compressive Embedding and Visualization using Graphs
Cited by in corpus (7)
- Prot2Text: Multimodal Protein's Function Generation with GNNs and Transformers
- What graph neural networks cannot learn: depth vs width
- Erdos Goes Neural: an Unsupervised Learning Framework for Combinatorial Optimization on Graphs
- Node Masking: Making Graph Neural Networks Generalize and Scale Better
- Quaternion Graph Neural Networks
- A Structural Feature-Based Approach for Comprehensive Graph Classification
- Weisfeiler and Leman go Hyperbolic: Learning Distance Preserving Node Representations