2021 Fiscal Year Final Research Report

Structured Convolutional Networks for High-dimensional Signal Restoration

Research Project

PDF

Project/Area Number	19H04135
Research Category	Grant-in-Aid for Scientific Research (B)
Allocation Type	Single-year Grants
Section	一般
Review Section	Basic Section 61010:Perceptual information processing-related
Research Institution	Niigata University
Principal Investigator	Muramatsu Shogo 新潟大学, 自然科学系, 教授 (30295472)
Co-Investigator(Kenkyū-buntansha)	日比野浩大阪大学, 医学系研究科, 教授 (70314317) 崔森悦新潟大学, 自然科学系, 准教授 (60568418) 山田寛喜新潟大学, 自然科学系, 教授 (20251788) 安田浩保新潟大学, 災害・復興科学研究所, 研究教授 (00399354) 湯川正裕慶應義塾大学, 理工学部(矢上), 准教授 (60462743) 小野峻佑東京工業大学, 情報理工学院, 准教授 (60752269)
Project Period (FY)	2019-04-01 – 2022-03-31
Keywords	スパースモデリング / 信号復元 / 信号推定 / 辞書学習 / 畳み込み構造
Outline of Final Research Achievements	The purpose of this study was to clarify effective structural constraints on generative models of high-dimensional signals in order to improve the efficiency and performance of high-dimensional signal restoration. New structured dictionaries and structured convolutional networks were constructed, and efforts were made to improve the efficiency of complex extension, parallelization, hierarchization, multilayering, learning design, and restoration algorithms. We evaluated the effectiveness of the proposed restoration process on a variety of real data, including volumetric data, complex image data, and high-dimensional time series data. As research results, we made 21 presentations at domestic conferences, 16 presentations at international conferences, and published 3 papers in academic journals.
Free Research Field	信号処理
Academic Significance and Societal Importance of the Research Achievements	本研究課題の基盤には，研究代表者・村松が研究課題「基盤研究(C) 多次元信号復元のための事例に基づく非分離冗長重複変換の設計と実現」(26420347, 2014～2016年度)で挙げた成果があった．フィルタバンクの多層化，階層化，複素拡張，非線形拡張など未解決課題に取り組んだ結果，多様な高次元信号の有力な生成モデルが得られた．特に，最適化理論に基づき観測過程と生成過程を切り分けてネットワーク構造に反映した点など，既存の畳み込みネットワークにはない特徴を実現できた．国内学会や国際会議での発表，学術論文の掲載を通じて，本研究課題の成果が幅広い分野の発展に寄与できることを示せた．