2022 Fiscal Year Final Research Report
Accelerating Inductive Logic Programming Using GPU
Project/Area Number |
19K11909
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 60050:Software-related
|
Research Institution | Tokyo University of Science |
Principal Investigator |
|
Project Period (FY) |
2019-04-01 – 2023-03-31
|
Keywords | 帰納論理プログラミング / Progol / GPU / RDBMS / SQL / 群知能 / 粒子群最適化 / PSO |
Outline of Final Research Achievements |
Inductive Logic Programming (ILP), which is one of explainable A.I.s., generates a hypothesis based on training data consisting of positive examples and negative examples. In the process generating the hypothesis, ILP search for the most suitable hypothesis while repeating generating hypothesis candidates and checking whether they deduct the positive examples and do not deduct the negative examples. In this project, we implement parallel checking of the examples on RDBMS with GPU execution through converting the checking into SQL operations. Furthermore, we have enabled the SQL operations to simultaneously check several hypothesis candidates, so that we have achieved making it six times faster through decreasing the overhead of RDBMS. Also, we have implemented the method that generates the hypothesis based on PSO that is one of swarm intelligence algorithms, so that we have achieved twice more speedup.
|
Free Research Field |
計算機科学
|
Academic Significance and Societal Importance of the Research Achievements |
現在広く利用されているAIの深層学習は,学習内容がわからないために,分析ツールとして利用することは困難である.これに対して,ILPは,その学習内容を完璧に説明でき,新しいサンプルに対する推論も,推論過程を確認することができる.本研究は,説明可能AIであるILPを,並列化とメタヒューリスティクス化によって高速化し,ビックデータに適用できるようにした.今後,本AIシステムを,がんの臨床データと遺伝子データに適用することによって,予後や再発予想を行えるモデルを作成する予定である.
|