| 研究実績の概要 |
We have finished the project of learning on large graphs. Overall, we have derived many methods to extract information to encode local structures of graphs for learning purposes. Specially, we have derived many graph neural network models to extract information of molecules in different situations: to represent fingerprints and mass spectra. This is not only to learn mass spectra as a graph, but also learning to associate mass spectra information with molecular fingerprints jointly. Another work is to predict drug properties in adverse side effects taking into account both drugs as molecular graphs and its interaction with PPI graph. These models take into account a graph among graphs (as nodes) to extract information for the nodes, which are also graphs themselves. We also have derived a method is to use graphs as regularizers on large general hypergraphs.
|
