Tensor Network Representation for Machine Learning: Theoretical Study and Algorithms Development

2023 Fiscal Year Final Research Report

• Project/Area Number: 20H04249
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Institute of Physical and Chemical Research
• Principal Investigator: ZHAO Qibin 国立研究開発法人理化学研究所, 革新知能統合研究センター, チームリーダー (30599618)
• Co-Investigator(Kenkyū-buntansha): 曹 建庭 埼玉工業大学, 工学部, 教授 (20306989) ; 横田 達也 名古屋工業大学, 工学(系)研究科(研究院), 准教授 (80733964)
• Project Period (FY): 2020-04-01 – 2024-03-31
• Keywords: tensor network / machine learning / adversarial robustness

Outline of Final Research Achievements

We studied and developed some advanced tensor decomposition and tensor network representation methods for incomplete and noisy data tensor. To improve its practicability, we also developed various algorithms for optimal tensor network structure search, and deep neural network based nonlinear flexible tensor decomposition methods. Moreover, we also studied adversarial robustness of deep neural networks under tensor representation of model parameters and developed several novel approaches for adversarial purification. Finally, our research findings can be adopted into some applications such as multi-model learning, hyperspectral image processing and etc.

Free Research Field

machine learning

Academic Significance and Societal Importance of the Research Achievements

Our research further promotes fundamental technology on tensor methods for machine learning. We have shown that tensor methods are powerful for structured data analysis and also practically useful for parameter representation of deep neural networks, resulting in more efficient and robust models.