High-Order Deep Learning Models: Theoretical Study and Applications

2019 Fiscal Year Final Research Report

• Project/Area Number: 17K00326
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• 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, ユニットリーダー (30599618)
• Co-Investigator(Kenkyū-buntansha): 曹 建庭 埼玉工業大学, 工学部, 教授 (20306989)
• Project Period (FY): 2017-04-01 – 2020-03-31
• Keywords: Tensor decomposition / Tensor networks / Multi-modal learning / Tensor completion

Outline of Final Research Achievements

Tensor decomposition and tensor networks (TNs) have recently gained increasing attentions in machine learning, data mining and computer vision fields due to its effectiveness in efficient computation and model compression in deep learning. However, there are many open problems that are still unexplored, which limits its impact in machine learning. In this project, we studied the fundamental model and theory of tensor networks and applied it for data representation and model representation. We have introduced a novel tensor decomposition model together with fast and scalable algorithms which can be applied to large-scale data completion and image denoising. In addition, we also developed deep multi-task, multi-model learning and multiple GANs methods based on tensor network representations, which shows powerful expressive ability and economic model complexity.

Free Research Field

Information Science

Academic Significance and Societal Importance of the Research Achievements

Tensor representation and tensor networks have shown to be useful in deep learning models. This project has further promoted to solve the challenging problems in deep learning methods by using TN technology.