Locally Structured Unitary Networks and Tangent Space Learning for Dynamic Systems Modeling

用于动态系统建模的局部结构酉网络和切线空间学习

• 批准号: 22H00512
• 负责人: 村松 正吾
• 金额: $ 27.21万
• 依托单位: Niigata University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (A)
• 财政年份: 2022
• 资助国家: 日本
• 起止时间: 2022-04-01 至 2026-03-31
• 项目状态: 未结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-22H00512/
• 关键词: 信号処理; 制御理論; 接空間学習; 動的システム; データ駆動モデリング; 次元削減

計測技術の発展と共に，複雑な物理現象の理解と予測，制御への要求が高まり，データ駆動によるモデル化，特に，畳み込みニューラルネットワーク(CNN)を利用した手法が有力視されてきた．しかし，CNNは解釈性に乏しく，領域知識をその構造に反映し難い課題がある．そこで本研究では，畳み込み層の局所性を継承し，ユニタリ性の構造制約の下で適応制御可能な線形層の実現を試みている．接空間に直接対応するその構造は高い解釈性を与え，現象の理解を促進し，その予測と制御に高性能化をもたらすことが期待できる．防災に重要な河床や地盤の変動を事例とし，情報学，土木工学，数学，物理学の融合研究体制で課題に取り組み，防災応用での有効性の実証を目指している．具体的には，動的システムのモデル化に有効な線形構造を探求している．新たに局所構造化ユニタリ層を導入し，汎用的で解釈性が高いデータ駆動による動的システムのモデリング手法の確立に取り組んでいる．2022年度は主に局所構造化ユニタリネットワーク(LSUN)の学習設計に取り組んだ．以下に成果をまとめる．(i) 開発: LSUNをソフトウェア実装し，データ駆動による動的システムのモデル化を行う環境を整えた．(ii) 解析：Burgers方程式に従う時系列データの解析をとおして，LSUNの性能評価を実施した．(iii) 実証: 河川の河道変動の実験や地盤の変動の観測から時系列データを取得し，畳み込み構造などLSUNに関連する動的システムモデリング手法の提案と評価を進め，LSUNの有効性の実証に必要なデータや手法の整理を行った．なお，提案手法の基礎的な評価には数値計算ソフトウェアMATLABを採用し，ソースコードを公開するための準備も進めた．適宜，国内学会，国際会議にて成果発表を行うと同時に，基礎的な研究成果の学術論文誌への投稿準備も進めた．

Measurement technology is developed and shared, complex physical phenomena are understood and predicted, control requirements are high, and movement is doneモデル化, special に, 畳み込みニューラルネットワーク (CNN) をUse the した technique and されてきた.しかし, CNN's unsolvable nature is lacking, and domain knowledge is structured to reflect difficult topics. The purpose of this study is to determine the local nature of the layer and the structural constraints of the layer, and to control the possibility of linear layering. The space is directly connected to the structure and structure, the understanding of the phenomenon is promoted, and the prediction and control are high-performance and expected. Disaster prevention is important, the river bed and the ground are moving, the case study is done, the integrated research system of informatics, civil engineering, mathematics, and physics is the project group, and the effectiveness of the disaster prevention and control application is confirmed. The concrete structure, dynamic structure, effective linear structure, and exploration of the structure. The introduction of the new structure's structural layer, the general-purpose high-performance solution The システムのモデリング technique of 駆动による动 is established and the group んでいる is established. In 2022, the main bureau's structured training program (LSUN)'s learning design group was selected. The following results are as follows: (i) Open: LSUN をソフトウェア実装し, データ駆动による动的システムのモデル化を行うEnvironment を全えた． (ii) Analysis: Burgers equation, time series analysis, LSUN performance evaluation. (iii) Certification: The river's river course is moving, the land is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the river is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the ground is moving, the river is moving, the river is moving, the ground is moving, the river is moving. Proposal and evaluation of the システムモデリング technique, と価をEnter, LSUN's effectiveness and proof, necessary なデータや technique, arrangement and execution.なお, the basis of the proposal method is the calculation of the numerical value, the MATLAB is used, the ソースコードを is published, and the preparation is carried out. Suitable for domestic societies and international conferences, the presentation of results is carried out and at the same time, basic research results are recorded and academic papers are prepared for submission.

项目成果

有向グラフ信号に対する実フィルタ設計に関する検討

有向图信号的实滤波器设计研究

• 发表时间: 2023
• 作者: 山賀俊介;竹内洋輝;天野敦雄;村松正吾，北村帆高，安田浩保，田中雄一
• 通讯作者: 村松正吾，北村帆高，安田浩保，田中雄一

Sparsity-Aware OCT Volumetric Data Restoration Using Optical Synthesis Model

• DOI: 10.1109/tci.2022.3183396
• 发表时间: 2022-01-01
• 期刊: IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING
• 影响因子: 5.4
• 作者: Kobayashi，Ruiki;Fujii，Genki;Muramatsu，Shogo
• 通讯作者: Muramatsu，Shogo

Graph Spatio-Spectral Total Variation Model for Hyperspectral Image Denoising

• DOI: 10.1109/lgrs.2022.3192912
• 发表时间: 2022-07
• 期刊: IEEE Geoscience and Remote Sensing Letters
• 影响因子: 4.8
• 作者: Shingo Takemoto;Kazuki Naganuma;Shunsuke Ono

GitHub - SaivDr Package for MATLAB/Simulink ホームページ

GitHub - MATLAB/Simulink 主页的 SaivDr 包

多次元信号・画像処理研究室 － 研究プロジェクト

多维信号与图像处理实验室-研究项目