2020 Fiscal Year Annual Research Report
Tensor Network Representation for Machine Learning: Theoretical Study and Algorithms Development
| Project/Area Number |
20H04249
|
|---|
| 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 | machine learning / tensor network |
|---|
| Outline of Annual Research Achievements |
In this year, our research has mainly addressed the problem of learning tensor network representation from data and deep learning modeling with tensor network fusion. Specifically, we have several contributions that are listed as follows.
1. We have developed reshuffled tensor decomposition and robust tensor Tubal nuclear norm based algorithm with theoretical support, which can provide exact recovery guarantee and improved tensor completion performance.
2. We developed outer product based tensor fusion framework, which can be employed in deep multimodal learning yielding ability to handing incomplete data. The experiments on multimodal sentiment analysis has validate its effectiveness and improved performance.
|
| Current Status of Research Progress |
Current Status of Research Progress
3: Progress in research has been slightly delayed.
Reason
The research progress is slightly delayed due to COVID-19 issues.
|
| Strategy for Future Research Activity |
In the next years, we plan to conduct research on tensor network representation and deep learning. The big model in deep learning is able to produce high performance but the storage and computation efficiency is low. To address this issue, we aim to develop effective model compression technology using tensor network representation, which can be applied to reduce significantly the number of parameters in modeling while keeping the performance comparable.
|
