| Presentation | 2022-11-30 Evaluation of Model Quantization Method on Vitis-AI for Mitigating Adversarial Examples Yuta Fukuda, Kota Yoshida, Takeshi Fujino, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Adversarial examples (AEs) are security threats in deep neural networks (DNNs). One of the countermeasures is adversarial training (AT), and it trains DNNs by using a training dataset containing AEs to achieve robustness against AEs. On the other hand, it has been reported that the robustness of AT is lost when it quantizes AT-trained model parameters from the commonly used 32-bit floating point to 8-bit integer number to run DNN on edge devices such as FPGA. In a previous study, we pointed out that the cause is in a fine-tuning process in the quantization method that uses natural samples to deal with quantization errors. We have proposed quantization-aware adversarial training (QAAT) to address the problem, which optimizes DNNs by conducting AT in quantization flow. In this paper, we construct a QAAT model using Vitis-AI provided by Xilinx. We actually run on the evaluation board ZCU104, which is equipped with Zynq UltraScale+, and we evaluate the robustness of a QAAT-trained model against AEs. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Adversarial examples / Adversarial training / Vitis-AI / FPGA |
| Paper # | VLD2022-51,ICD2022-68,DC2022-67,RECONF2022-74 |
| Date of Issue | 2022-11-21 (VLD, ICD, DC, RECONF) |
| Conference Information | |
| Committee | VLD / DC / RECONF / ICD / IPSJ-SLDM |
|---|---|
| Conference Date | 2022/11/28(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Kanazawa Bunka Hall |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Design Gaia 2022 -New Field of VLSI Design- |
| Chair | Minako Ikeda(NTT) / Tatsuhiro Tsuchiya(Osaka Univ.) / Kentaro Sano(RIKEN) / Masafumi Takahashi(Kioxia) / Hiroyuki Ochi(Ritsumeikan Univ.) |
| Vice Chair | Shigetoshi Nakatake(Univ. of Kitakyushu) / Toshinori Hosokawa(Nihon Univ.) / Yoshiki Yamaguchi(Tsukuba Univ.) / Tomonori Izumi(Ritsumeikan Univ.) / Makoto Ikeda(Univ. of Tokyo) |
| Secretary | Shigetoshi Nakatake(NBS) / Toshinori Hosokawa(Hirosaki Univ.) / Yoshiki Yamaguchi(Nihon Univ.) / Tomonori Izumi(Chiba Univ.) / Makoto Ikeda(NEC) / (Toyohashi Univ. of Tech.) |
| Assistant | Takuma Nishimoto(Hitachi) / / Yukitaka Takemura(INTEL) / Yasunori Osana(Ryukyu Univ.) / Yoshiaki Yoshihara(KIOXIA) / Jun Shiomi(Osaka Univ.) / Takeshi Kuboki(Sony Semiconductor Solutions) |
| Paper Information | |
| Registration To | Technical Committee on VLSI Design Technologies / Technical Committee on Dependable Computing / Technical Committee on Reconfigurable Systems / Technical Committee on Integrated Circuits and Devices / Special Interest Group on System and LSI Design Methodology |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Evaluation of Model Quantization Method on Vitis-AI for Mitigating Adversarial Examples |
| Sub Title (in English) | |
| Keyword(1) | Adversarial examples |
| Keyword(2) | Adversarial training |
| Keyword(3) | Vitis-AI |
| Keyword(4) | FPGA |
| 1st Author's Name | Yuta Fukuda |
| 1st Author's Affiliation | Ritsumeikan University(Ritsumeikan Univ.) |
| 2nd Author's Name | Kota Yoshida |
| 2nd Author's Affiliation | Ritsumeikan University(Ritsumeikan Univ.) |
| 3rd Author's Name | Takeshi Fujino |
| 3rd Author's Affiliation | Ritsumeikan University(Ritsumeikan Univ.) |
| Date | 2022-11-30 |
| Paper # | VLD2022-51,ICD2022-68,DC2022-67,RECONF2022-74 |
| Volume (vol) | vol.122 |
| Number (no) | VLD-283,ICD-284,DC-285,RECONF-286 |
| Page | pp.pp.182-187(VLD), pp.182-187(ICD), pp.182-187(DC), pp.182-187(RECONF), |
| #Pages | 6 |
| Date of Issue | 2022-11-21 (VLD, ICD, DC, RECONF) |