Constructive approach to investigate the emergent mechanisms of visual and sensory function

2021 Fiscal Year Final Research Report

• Project/Area Number: 19H04200
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61060:Kansei informatics-related
• Research Institution: National Institute of Advanced Industrial Science and Technology
• Principal Investigator: Hayashi Ryusuke 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (80444470)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 深層学習 / 視覚情報処理 / 感性情報処理

Outline of Final Research Achievements

In this study, we developed a deep neural network that acquires a representation style similar to the brain's visual information representation based on the knowledge of visual information processing in the brain, learning principles, information processing constraints, and network structure. We analyzed the deep neural network from the viewpoint of sensory informatics, and measured and verified functional brain activity data. Specifically, we analyzed 1) neural networks for video processing, 2) analyzed visual representations using adversarial generative neural networks (GAN), which is an unsupervised learning framework, and developed them into hardware implementations, and 3) examined fNIRS measurements, neural recordings using multi-point electrodes. We also investigated 4) conceptual information representations in visual, language, and brain information. The research results were presented at conferences and published in papers.

Free Research Field

視覚科学

Academic Significance and Societal Importance of the Research Achievements

本研究成果で得られた深層学習フレームワークに関する知見は、脳が符号化する視覚情報表現と近い情報表現の機械学習による獲得手法に関して新たな知見を提供した点で学術的意義があると考える。将来、記録した神経活動データから、視覚体験を可視化するブレイン・マシン・インタフェース技術のような、工学応用を通して、より社会の実現に貢献できると考える。