• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to project page

2018 Fiscal Year Research-status Report

High-Order Deep Learning Models: Theoretical Study and Applications

Research Project

Project/Area Number 17K00326
Research InstitutionInstitute 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
KeywordsTensor ring / Deep multi-task learning
Outline of Annual Research Achievements

We studied multi-task deep learning models by using high-order tensor network technology. The existing multi-task deep learning models mostly based on sharing the lower layers and having individual upper layers for each task. The problem is that such framework cannot handle the case when different tasks have different input dimensions and when different tasks have heterogenous network architecture. To solve this challenging problem, we proposed a new framework, which allows us to use heterogenous network architecture for individual tasks, by leveraging tensor ring representation of weight parameters of each layer and some of latent core tensors sharing between tasks. The experiments demonstrated that our method is more flexible with high performance.

We studied tensor ring decomposition for data completion and developed several different methods to reconstruct the original tensor. The results show that tensor ring model is more powerful than traditional tensor decomposition models. We also give the theoretical support for our proposed methods.

We propose a new type of tensor decomposition to find the latent low-rank tensors under reshuffling operations. This is quite useful for image steganography. And the theoretical analysis shows the uniqueness condition of our new algorithm.

Current Status of Research Progress
Current Status of Research Progress

2: Research has progressed on the whole more than it was originally planned.

Reason

The project is performing smoothly.

Strategy for Future Research Activity

In this fiscal year, we plan to conduct following two research works.

1. The current CNN for 3D data like video clip needs computation cost that is exponentially with dimension of data, which is a challenging problem if the data is higher dimensions much larger than 3. We attempt to solve this problem by using tensor network representation to improve the computation cost to be linear with the dimension of data.

2. We will also study how to learning an optimal tensor network structure from given data. This is an open problem in tensor research field.

Causes of Carryover

The conference paper accepted in 2018 will be held in 2019.

  • Research Products

    (9 results)

All 2019 2018

All Journal Article (6 results) (of which Int'l Joint Research: 3 results,  Peer Reviewed: 6 results,  Open Access: 2 results) Presentation (3 results) (of which Int'l Joint Research: 2 results)

  • [Journal Article] High-order tensor completion via gradient-based optimization under tensor train format2019

    • Author(s)
      Yuan Longhao、Zhao Qibin、Gui Lihua、Cao Jianting
    • Journal Title

      Signal Processing: Image Communication

      Volume: 73 Pages: 53~61

    • DOI

      https://doi.org/10.1016/j.image.2018.11.012

    • Peer Reviewed
  • [Journal Article] Image and video completion by using Bayesian tensor decomposition2018

    • Author(s)
      L. Gui, X. Zhao, Q. Zhao, and J. Cao.
    • Journal Title

      International Journal of Computer Science Issues

      Volume: 15 Pages: 1~8

    • DOI

      https://doi.org/10.5281/zenodo.1467644

    • Peer Reviewed / Open Access
  • [Journal Article] Non-local image denoising by using Bayesian low- rank tensor factorization on high-order patches2018

    • Author(s)
      L. Gui, X. Zhao, Q. Zhao, and J. Cao.
    • Journal Title

      International Journal of Computer Science Issues

      Volume: 15 Pages: 16~25

    • DOI

      https://doi.org/10.5281/zenodo.1467648

    • Peer Reviewed / Open Access
  • [Journal Article] Feature Extraction for Incomplete Data Via Low-Rank Tensor Decomposition With Feature Regularization2018

    • Author(s)
      Shi Qiquan、Cheung Yiu-Ming、Zhao Qibin、Lu Haiping
    • Journal Title

      IEEE Transactions on Neural Networks and Learning Systems

      Volume: x Pages: 1~15

    • DOI

      10.1109/TNNLS.2018.2873655

    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Brain-Computer Interface with Corrupted EEG Data: a Tensor Completion Approach2018

    • Author(s)
      Sole-Casals J.、Caiafa C. F.、Zhao Q.、Cichocki A.
    • Journal Title

      Cognitive Computation

      Volume: 10 Pages: 1062~1074

    • DOI

      https://doi.org/10.1007/s12559-018-9574-9

    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Two-Stage Frequency Recognition Method Based on Correlated Component Analysis for SSVEP-Based BCI2018

    • Author(s)
      Zhang Yangsong、Yin Erwei、Li Fali、Zhang Yu、Tanaka Toshihisa、Zhao Qibin、Cui Yan、Xu Peng、Yao Dezhong、Guo Daqing
    • Journal Title

      IEEE Transactions on Neural Systems and Rehabilitation Engineering

      Volume: 26 Pages: 1314~1323

    • DOI

      10.1109/TNSRE.2018.2848222

    • Peer Reviewed / Int'l Joint Research
  • [Presentation] Higher-dimension tensor completion via low-rank tensor ring decomposition2018

    • Author(s)
      L. Yuan, J. Cao, X. Zhao, Q. Wu, and Q. Zhao.
    • Organizer
      APSIPA-ASC 2018
    • Int'l Joint Research
  • [Presentation] Epileptic focus localization based on iEEG by using positive unlabeled (PU) learning2018

    • Author(s)
      X. Zhao, T. Tanaka, W. Kong, Q. Zhao, J. Cao, H. Sugano, and N. Yoshida
    • Organizer
      APSIPA-ASC
    • Int'l Joint Research
  • [Presentation] Detection of Epileptic Foci Based on Interictal iEEG by Using Convolutional Neural Network2018

    • Author(s)
      Xuyang Zhao, Qibin Zhao, Toshihisa Tanaka, Jianting Cao, Wanzeng Kong, Hidenori Sugano and Noboru Yoshida∥
    • Organizer
      International Conference on Digital Signal Processing

URL: 

Published: 2019-12-27  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi