2020 Fiscal Year Research-status Report
Research on Learning Graphs via Enumerative Queries and its Applications
| Project/Area Number |
20K11998
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
Parque Victor 早稲田大学, 理工学術院, 准教授(任期付) (50745221)
|
|---|
| Project Period (FY) |
2020-04-01 – 2023-03-31
|
| Keywords | graph learning / planning / succinct representation / robotics / optimization |
|---|
| Outline of Annual Research Achievements |
The foundations for learning and optimization of graphs via enumerative queries and a number of application benchmarks in computer-aided design and robotics were proposed: (1) the efficient sampling algorithms for directed graphs and the rendering of Steiner trees on the plane with obstacles, (2) a novel sampling-based differential particle scheme for nonlinear optimization and control problems, (3) robust sampling algorithms using enumerative queries and efficient parallel reductions on the Graphics Processing Unit (GPU) to tackle combinatorial problems, which is crucial to learn directed graphs, undirected graphs, graphs with self-loops (loopy-graphs), and graphs with modules (hypergraphs), (4) new representations and stochastic optimization algorithms to compute collision-free lattice paths on binary occupancy maps with utmost efficiency, in about 10 ms., (5) a new scheme to learn motion planning functions in 6-DOF robot manipulators using the linear transition in the C-space, whose effectiveness has been demonstrated by Neural Nets and Kernel Machines, (6) the representations of higher order fairness functionals in curves to enable smooth and safe path planning for multi-agent systems, and (7) the new schemes for collision-free navigation of multi-agent systems in confined environments, providing the state-of-the art efficiency in computing safe navigation paths.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The key foundations and benchmarks on how to represent, learn and optimize graphs via enumerative queries network design and robotics problems were established.
|
| Strategy for Future Research Activity |
In 2021, while gaining awareness and insights toward the deployment in real-world systems and extending the basic algorithms, the research on the application of Learning Graphs with Enumerative Queries is to be conducted. The application in architecture search of Convolutional Neural Networks (tensor decompositions), and the optimal search of Bayesian Network Structure (combinatorial decompositions) are to be conducted. Also, the application in re al-world Multi-Agent Systems (mobile robots) to enable safe and smooth path planning is to be conducted. And, research on the de sign of sensor networks for applications in Industrial Internet of Things (IIoT), Agriculture, and Body Sensor Networks are to be conducted.
|
| Causes of Carryover |
The reasons for incurring the amount to be used in the next fiscal year are as follows: (1) Regarding Article Costs, savings are due to postponing the acquisition of a computer for numerical computations to the next fiscal year. (2) Regarding Travel Expenses, the savings are due to attending virtual conferences (pandemic situation). (3) Regarding miscellaneous costs, the journal articles corresponding to the extensions/developments of the ideas presented at conferences in the fiscal year 2020 are under submission/preparation. The amount to be used in the next fiscal year will consider the costs for acquiring a computing environment suitable for computationally expensive and concurrent calculations and the costs for publishing at relevant conference and journal venues.
|
