2020 Fiscal Year Research-status Report
Learning Internal Representations Robust against Adversarial Attacks
| Project/Area Number |
20K19824
|
|---|
| Research Institution | Kyushu University |
|---|
Principal Investigator |
VARGAS DANILO 九州大学, システム情報科学研究院, 准教授 (00795536)
|
|---|
| Project Period (FY) |
2020-04-01 – 2022-03-31
|
| Keywords | 敵対的機械学習 / Deep Neural Networks / 深層学習 / One pixel attack |
|---|
| Outline of Annual Research Achievements |
The planned architecture was developed and tested. The improvement of the robustness was also verified. Variations of the architecture using generative neural networks and autoencoders were realized and tested. It was also proposed new methods to evaluate the latent space of variables in both autoencoders and layers of deep neural networks. All these results were described in detail and submitted/published in papers.
Regarding the research achievements, 7 papers in international conferences and 1 book was published. Additionally, there were an invited talk about the subject and 2 tutorials in top conferences such as IJCAI and WCCI.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
Research experiments could be finished without great problems, leading to a more smooth than initially planned status. Currently most of the research was/is being submitted to top conferences.
|
| Strategy for Future Research Activity |
This year beyond finishing the last part of the planned research, other types of robust architectures/paradigms for computer vision will be also investigated.
|
| Causes of Carryover |
COVID-19の影響に応じて、必要でした。
|
