Training Free Graph Neural Networks for Graph Matching. (arXiv:2201.05349v1 [cs.LG])

We present TFGM (Training Free Graph Matching), a framework to boost the
performance of Graph Neural Networks (GNNs) based graph matching without
training. TFGM sidesteps two crucial problems when training GNNs: 1) the
limited supervision due to expensive annotation, and 2) training's
computational cost. A basic framework, BasicTFGM, is first proposed by adopting
the inference stage of graph matching methods. Our analysis shows that the
BasicTFGM is a linear relaxation to the quadratic assignment formulation of
graph matching. This guarantees …

arxiv graph graph neural networks networks neural networks training