Parallel Search Algorithms for Heterogeneous Computing Environments

• Project/Area Number: 17K00296
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: The University of Tokyo
• Principal Investigator: Fukunaga Alex 東京大学, 大学院総合文化研究科, 教授 (90452002)
• Project Period (FY): 2017-04-01 – 2021-03-31
• Project Status: Completed (Fiscal Year 2020)
• Budget Amount: ¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000); Fiscal Year 2019: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000); Fiscal Year 2018: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000); Fiscal Year 2017: ¥1,820,000 (Direct Cost: ¥1,400,000、Indirect Cost: ¥420,000)
• Keywords: 探索 / 並列アルゴリズム / Heuristic search / artificial intelligence / 探索アルゴリズム / 人工知能 / 並列探索アルゴリズム

Outline of Final Research Achievements

We developed Block-parallel IDA*, an admissible search algorithm for finding optimal solutions for heterogeneous CPU/GPU environments. Block-parallel IDA* allocates subtrees of a search tree to each GPU block. We showed that on the standard sliding tiles benchmark domain, Block-parallel IDA* significantly outperforms thread-parallel IDA*.
We also developed Bach Monte Carlo Random Walk, a satisficing (non-optimal) search algorithm which significantly outperformed previous Random Walk search algorithms on a large set of domain-independent planning benchmark instances.

Academic Significance and Societal Importance of the Research Achievements

近年、従来型の演算コアを数十個搭載する主演算装置（CPU）に加えて、数千個の演算コアを搭載するGPUを両方搭載しているヘテロジニアスな計算機が普及している。人工知能においてエージェントやロボットの自動行動計画問題及び、経営工学における生産スケジューリング問題や施設配置問題等、多くの難解な探索・最適化問題に対して並列処理能力を十分に発揮できるグラフ探索アルゴリズムを設計することが必要である。本研究においてCPUとGPUを同時に有効に使う探索アルゴリズムを提案・解析した。

Research Products

• [Journal Article] On the Pathological Search Behavior of Distributed Greedy Best-First Search (2019)
  • Author(s): Ryo Kuroiwa, Alex Fukunaga
  • Journal Title: Proceedings of the International Conference on Automated Planning and Scheduling Volume: 1 Pages: 255-263
  • Peer Reviewed / Open Access
• [Journal Article] Batch Random Walk for GPU-Based Classical Planning (2018)
  • Author(s): Ryo Kuroiwa, Alex Fukunaga
  • Journal Title: Proceedings of the International Conference on Automated Planning and Scheduling Volume: 1 Pages: 155-160
  • Peer Reviewed / Open Access
• [Journal Article] Block-Parallel IDA* for GPUs (2017)
  • Author(s): Satoru Horie, Alex Fukunaga
  • Journal Title: Proceedings of the Symposium on Combinatorial Search Volume: 1 Pages: 1-8
  • Peer Reviewed / Open Access
• [Presentation] Learning Search-Space Specific Heuristics Using Neural Networks (2020)
  • Author(s): Liu Yu, Ryo Kuroiwa, Alex Fukunaga
  • Organizer: 12th Workshop on Heuristics and Search for Domain-Independent Planning
  • Int'l Joint Research