2015 Fiscal Year Annual Research Report
Fast Optimal Transport and Applications to Inference and Simulation in Large Scale Statistical Machine Learning
| Project/Area Number |
26700002
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
Cuturi Marco 京都大学, 情報学研究科, 准教授 (80597344)
|
|---|
| Project Period (FY) |
2014-04-01 – 2017-03-31
|
| Keywords | 統計的機械学習 / 最適輸送理論 / 最適化 / グラフィックス |
|---|
| Outline of Annual Research Achievements |
The purpose of this research is to explore the novel possibilities that optimal transport theory can provide to statistical modeling, machine learning and other related fields such a graphics and optimization. To do so, the principal investigator of this grand has proposed in 2013 a key result which allows for the resolution of the optimal transport using massively parallel architectures such as GPGPU. This allowed me to propose, within the framework of this project, several ideas in FY2014 which led to notable publications. FY2015 produced even more significant results for our research project. The most noteworthy of these results were the publications of 2 papers in SIAM Journals (SIAM Journal on Scientific Computing, SIAM Journal on Imaging Analysis), coupled with publications in NIPS and SIGGRAPH, arguably the two most visible conferences in computer science.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The excellence of the publications we have produced this year clearly demonstrate the importance of our results. We expect several more achievements in coming months, both from machine learning and/or graphics. We are also planning several scientific events later this year and in 2017.
|
| Strategy for Future Research Activity |
We are now consolidating the project, and will therefore aim at producing summaries, reviews, and any other material that might provide additional visibility to our work. Our goal will thus include the publication of a book and the continuation of our publishing strategy, in which we aim for the best possible conferences and venues to give maximal publicity to the topic of modern / numerical optimal transport.
|
| Causes of Carryover |
The program is reaching its final year. We need, more than ever, to connect with other researchers in the world and publicize our findings. Our aim in this final year will be in particular to disseminate these new methods in other fields, such applied/numerical mathematics, statistics and physics.
|
| Expenditure Plan for Carryover Budget |
The biggest source of expense will come from travel. We expect expenses coming from participation to conferences and other forums (workshops, research visits), both for the principal investigator of this project and collaborators (including students). We also expect to keep on spending a small fraction of our budget on hardware (to add to our computational resources) and smaller machines.
|
