A Prufer-Sequence Based Representation of Large Graphs for Structural Encoding of Logic Networks
This work addresses the need for efficient graph representations in domains like VLSI circuit analysis, though it appears incremental as it extends tree-based encoding to general graphs.
The paper tackles the problem of representing large graphs for machine learning by proposing a lossless, linear-sized 1-D encoding based on Prufer sequences, which achieved a representation that scales with the number of vertices and includes graph properties for interpretability.
The pervasiveness of graphs in today's real life systems is quite evident, where the system either explicitly exists as graph or can be readily modelled as one. Such graphical structure is thus a store house rich information. This has various implication depending on whether we are interested in a node or the graph as a whole. In this paper, we are primarily concerned with the later, that is, the inference that the structure of the graph influences the property of the real life system it represents. A model of such structural influence would be useful in inferencing useful properties of complex and large systems, like VLSI circuits, through its structural property. However, before we can apply some machine learning (ML) based technique to model such relationship, an effective representation of the graph is imperative. In this paper, we propose a graph representation which is lossless, linear-sized in terms of number of vertices and gives a 1-D representation of the graph. Our representation is based on Prufer encoding for trees. Moreover, our method is based on a novel technique, called $\mathcal{GT}$-enhancement whereby we first transform the graph such that it can be represented by a singular tree. The encoding also provides scope to include additional graph property and improve the interpretability of the code.