| Presentation | 2019-03-13 Improving the robusteness of neural networks to adversarial examples by reducing color depth of training inage data Shuntaro Miyazato, Toshihiko Yamasaki, Kiyoharu Aizawa, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In this research, we propose a method to train a neural network that is robust to adversarial examples to image classification. First, we show that the accuracy of adversarial example can be improved while keeping the accuracy of data which is not adversarial example by dropping RGB value information of the training image. In addition, we suggest that the robustness improves further by determining the quantization level so that the loss function is maximized just before back propagation of the neural network. Finally, we report the ensemble of the model trained with quantization accomplished the same performance as the model adversarial trained, if they can reject indeterminate examples. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | deep learning / adversarial example |
| Paper # | EMM2018-109 |
| Date of Issue | 2019-03-06 (EMM) |
| Conference Information | |
| Committee | EMM |
|---|---|
| Conference Date | 2019/3/13(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | TBD |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Image and Sound Quality, Metrics for Perception and Recognition, Human Auditory and Visual System, etc. |
| Chair | Keiichi Iwamura(TUC) |
| Vice Chair | Minoru Kuribayashi(Okayama Univ.) / Tetsuya Kojima(NIT,Tokyo College) |
| Secretary | Minoru Kuribayashi(NIT, Tokyo) / Tetsuya Kojima(Tyukyo Univ.) |
| Assistant | Hiroko Akiyama(NIT, Nagano College) / キタヒロ カネダ(CANON) |
| Paper Information | |
| Registration To | Technical Committee on Enriched MultiMedia |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Improving the robusteness of neural networks to adversarial examples by reducing color depth of training inage data |
| Sub Title (in English) | |
| Keyword(1) | deep learning |
| Keyword(2) | adversarial example |
| 1st Author's Name | Shuntaro Miyazato |
| 1st Author's Affiliation | The University of Tokyo(UTokyo) |
| 2nd Author's Name | Toshihiko Yamasaki |
| 2nd Author's Affiliation | The University of Tokyo(UTokyo) |
| 3rd Author's Name | Kiyoharu Aizawa |
| 3rd Author's Affiliation | The University of Tokyo(UTokyo) |
| Date | 2019-03-13 |
| Paper # | EMM2018-109 |
| Volume (vol) | vol.118 |
| Number (no) | EMM-494 |
| Page | pp.pp.95-100(EMM), |
| #Pages | 6 |
| Date of Issue | 2019-03-06 (EMM) |