Development of fast machine learning methods based on combinations of different computational models

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K11882
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: The University of Tokyo
• Principal Investigator: TANAKA GOUHEI 東京大学, ニューロインテリジェンス国際研究機構, 特任准教授 (90444075)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 機械学習 / 時系列データ / リザバー計算 / 非線形システム / 人工知能 / IoT社会

Outline of Final Research Achievements

To develop machine learning models capable of fast learning for temporal information processing, we constructed advanced machine learning models by employing the reservoir computing framework and evaluated the computational performance of the proposed models. We proposed advanced reservoir computing models through an introduction of feature extractions with resampling and filtering, an expansion of multi-reservoir computing models, a utilization of online learning methods inspired by transfer learning, and an exploitation of a reservoir consisting of heterogeneous computational units, and then demonstrated that the proposed methods are effective for enhancement of the computational performance and/or improvement in computational efficiency.

Free Research Field

複雑系動力学

Academic Significance and Societal Importance of the Research Achievements

本研究では、主にリザバー計算と他の計算手法の融合による高速機械学習モデルの開発を行った。異種計算モデルの融合に基づく高速機械学習手法は計算性能や計算効率の向上に有用であることが分かった。開発したモデルは、頻繁に学習計算をし直す必要のある環境や計算資源に制約がある環境における計算技術として有望であり、IoT社会においてエッジコンピューティング用の人工知能の基礎になると期待される。