Contents |
Graphs are widely considered as adequate forms of mathematical structures for expressing a wide class of non-Euclidean data, such as chemical compounds, genome sequences, trading networks, and social networking systems. The procedure of representing graphs as real vectors is a challenging problem indispensable to classifying innate properties of graphs. We explore a few strategies commonly used for obtaining real representations of graphs, which are Weisfeiler-Lehman tests, graph convolutional networks, and persistent homological techniques, along with assessments on their qualitative performances in classifying graph datasets. If time allows, we will discuss some possible research directions in addressing the limitations of these previously studied methods. |