Development of deep neural network architecture for multitask learning
| Project/Area Number |
18K11348
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61010:Perceptual information processing-related
|
|---|
| Research Institution | Tokyo Institute of Technology (2020)
The University of Tokyo (2018-2019) |
|---|
Principal Investigator |
Kawakami Rei 東京工業大学, 情報理工学院, 特任准教授 (90591305)
|
|---|
| Project Period (FY) |
2018-04-01 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥4,420,000 (Direct Cost: ¥3,400,000、Indirect Cost: ¥1,020,000)
Fiscal Year 2020: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2019: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2018: ¥2,210,000 (Direct Cost: ¥1,700,000、Indirect Cost: ¥510,000)
|
|---|
| Keywords | 深層学習 / マルチタスク学習 / ニューラルネットワーク / 物体検出 / 物体追跡 / 領域分割 / 汎化性能 / MTL |
|---|
| Outline of Final Research Achievements |
In this project, the applicant has developed a deep learner that performs multitask learning (MTL, Multitask Learning), in particular, MTL for the combination of object detection and semantic segmentation, and object tracking and detection. We have selected task combinations and data sets, proposed and improved cross-connections to realize MTL, and confirmed the improvement of generalization performance by MTL. The convolution of the cross-connections was replaced by a convolutional recurrent neural network, and a network for processing time-series data was designed, achieving performance improvement over conventional methods.
|
| Academic Significance and Societal Importance of the Research Achievements |
マルチタスク学習を行う深層学習器は今後，人に変わって認識を行うAIにおいて広く普及することになるだろう．本事業はそのマルチタスクの基本的な組み合わせにおいて，設計を吟味し，全般的な性能改善と汎化性能の向上に関する先駆的な結果を得た．一般的にマルチタスク学習では組み合わせても結果が向上するとは限らず，世界的にも様々な試みが提案されている中で，先駆的な結果を得たことは学術的にも社会的にも有意義である．
|
Report
(4 results)
Research Products
(28 results)
