S. Navuluru

Unifying Graph Neural Networks Through a Common Layer Equation

A common layer equation is introduced that represents covered architectures through seven components: an update domain, channel set, propagation bank, per-channel message maps, channel-fusion operator, ego/residual map, and update map, which exposes the empirical inverse problem of mapping measurable graph and task properties to validated component choices.