| Project/Area Number |
20K11932
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61030:Intelligent informatics-related
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
福永 ALEX 東京大学, 大学院総合文化研究科, 教授 (90452002)
|
|---|
| Project Period (FY) |
2020-04-01 – 2025-03-31
|
| Project Status |
Granted (Fiscal Year 2023)
|
| Budget Amount *help |
¥4,290,000 (Direct Cost: ¥3,300,000、Indirect Cost: ¥990,000)
Fiscal Year 2022: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2021: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2020: ¥1,950,000 (Direct Cost: ¥1,500,000、Indirect Cost: ¥450,000)
|
|---|
| Keywords | 探索アルゴリズム / 探索 / 人工知能 / 並列アルゴリズム / Heuristic Search |
|---|
| Outline of Research at the Start |
ロボット等のエージェントの自動行動計画等においてグラフ探索アルゴリズムが広く応用されている。大規模な問題を限られた時間内に解くには探索アルゴリズムの並列化が必要である。最短経路を求めるアルゴリズムについてはある程度効率良い並列化手法が提案されている。一方、大規模な問題の場合、最短経路を求めるのは困難な為、GBFS等、限られた時間内でなるべく良い経路を求めるアルゴリズムが使用されている。GBFSの効果的な並列化手法はアンサンブル法による以外、開発されていない。本研究ではGBFSの効率的な並列化手法の開発及び理論的解析を試みる。
|
| Outline of Annual Research Achievements |
In 2023-2024, we researched improved algorithms for parallel search. While parallelization of the A* graph search algorithm is fairly wellunderstood, parallelization of non-optimal best-first search algorithms such as Greedy Best-First Search (GBFS) has been much less understood. Recent theoretical work by Heusner, Keller, and Helmert (2017) identified the Bench Transition System (BTS), which is the set of states that can be expanded by GBFS under some tie-breaking policy.
In this project, we have been investigating a new class of algorithms which constrains parallel search such that only states which are guaranteed to be in the BTS are expanded.
This class of Constrained Parallel Greedy Best First Search (CPGBFS) algorithms include Parallel Under High-water mark First (PUHF) and its variants (PUHF2-4).
CPGBFS algorithms has a significantly slower state expansion rate than non-constrained parallel GBFS because threads spend much of the time waiting for the availability of states which satisfy the expansion constraints.
We developed Separate Generation and Evaluation (SGE), which decouples state sucessor generation and state evaluation, allowing more efficient usage of available threads.
|
| 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 goals of this project were (1) to analyze previously proposed parallel greedy best first search in order to understand how the behavior of parallel GBFS diverged from sequential GBFS, and (2) apply the theoretical insights obtained from (1) in order to develop new parallel GBFS algorithms which outperformed previous parallel GBFS strategies.
With regards to goal (1), our results published in (Kuroiwa and Fukunaga, 2020) showed that the behavior of previous parallel GBFS algorithms
could diverge arbitrarily from sequential GBFS. More specifically, previous parallel GBFS algorithms could not be guaranteed to search no more
than K times the nodes searched by sequential GBFS (for some constant K). Furthermore, it was shown that previous parallel GBFS algorithms expanded nodes which are not included the BTS, the set of expanded by sequential GBFS algorithms under some tie-breaking strategy.
Regarding goal (2), we proposed PUHF, a new parallel GBFS which is guaranteed to only expand nodes in the BTS (Kuroiwa and Fukunaga 2020), and proposed improvements to PUHF which improved upon the expansion criteria (Shimoda and Fukunaga 2023). Furthermore, in 2023-2024, we developed Separate Generation and Evaluation, which significantly improves the state evaluation rates of the PUHF-family of algorithms. Thus, we believe the project is achieving the goals set forth in the project proposal.
|
| Strategy for Future Research Activity |
In 2024-2025, we will complete the research project by completing the experimental evaluation of Separate Generation and Evaluation applied to the PUHF family of constrained parallel GBFS algorithms which we developed in 2023-2024.
We will complete a paper on SGE, and present the results at an international workshop.
|
