Graph Neural Networks for Learning Equivariant Representations of Neural Networks

Representing neural networks as computational graphs enables learning from diverse architectures using graph neural networks and transformers.