2018 Fiscal Year Research-status Report
Machine Learning on Large Graphs
| Project/Area Number |
18K11434
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2018-04-01 – 2021-03-31
|
| Keywords | Learning on graphs |
|---|
| Outline of Annual Research Achievements |
The target of this research is to learn from large graphs in a mathematically sound and practically efficient manner. For similarity graph, previous approaches are mainly based on graph Laplacians, which are theoretically not reliable, or graph p-Laplacians, which are computationally inefficient. We are considering the approaches based on nodes, flows and random walks. We made progress with random walks for general graphs that learn to encode structural information of graphs. This is an application of random walks on learning the structures of molecular graphs.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The main target, learning the structure of a large network is still in progress. We have set out the idea and general formulations. However, the true parameters of the formulations are still missing. We are looking for mathematical tools that could prove the correctness of the parameters. The progress of the project is subject to the width of the mathematical tools, which are now under exploration.
|
| Strategy for Future Research Activity |
We plan to continue our main target, looking for the right mathematical tools to prove the merit of our formulation. In parallel, we are applying these basic ideas into various challenging problem in machine learning such are encoding large patterns on graphs and clustering on graphs.
|
| Causes of Carryover |
The incurring amount to be used next year are mainly due to the reschedule of business trips and purchase of not yet necessary items. We plan to use them this year in the same way as planned for fiscal last year.
|
