| Presentation | 2022-09-08 Proposal and evaluation of Combined Posit MAC unit (CPMAC) for both DNN inference and training Yuta Masuda, Yasuhiro Nakahara, Masato Kiyama, Masahiro Iida, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Recently, there has been a lot of research on DNN hardware accelerators for the edge that use Posit as a number representation. Although the Posit contributes highly accurate inference and training using a fewer bit-width than floating point numbers, there is a difference in the required bits accuracy between inference and training, requiring other arithmetic units for each. However, it is difficult to implement both arithmetic units on edge devices with limited resources. In this article, we propose a Combined Posit MAC unit (CPMAC) for inference and training which can combine a lower precision Posit MACunit with the plural. As a result, we achieved an area reduction of more than 20% at the maximum when the exponent of Posit is large, and demonstrated the usefulness of CPMAC in applications that require a wide dynamic range. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | DeeoLearning / Convolutional Neural Network / Posit / MAC unit |
| Paper # | RECONF2022-34 |
| Date of Issue | 2022-08-31 (RECONF) |
| Conference Information | |
| Committee | RECONF |
|---|---|
| Conference Date | 2022/9/7(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | emCAMPUS STUDIO |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Reconfigurable system, etc. |
| Chair | Kentaro Sano(RIKEN) |
| Vice Chair | Yoshiki Yamaguchi(Tsukuba Univ.) / Tomonori Izumi(Ritsumeikan Univ.) |
| Secretary | Yoshiki Yamaguchi(NEC) / Tomonori Izumi(Toyohashi Univ. of Tech.) |
| Assistant | Yukitaka Takemura(INTEL) / Yasunori Osana(Ryukyu Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Reconfigurable Systems |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Proposal and evaluation of Combined Posit MAC unit (CPMAC) for both DNN inference and training |
| Sub Title (in English) | |
| Keyword(1) | DeeoLearning |
| Keyword(2) | Convolutional Neural Network |
| Keyword(3) | Posit |
| Keyword(4) | MAC unit |
| 1st Author's Name | Yuta Masuda |
| 1st Author's Affiliation | Kumamoto University(Kumamoto Univ.) |
| 2nd Author's Name | Yasuhiro Nakahara |
| 2nd Author's Affiliation | Kumamoto University(Kumamoto Univ.) |
| 3rd Author's Name | Masato Kiyama |
| 3rd Author's Affiliation | Kumamoto University(Kumamoto Univ.) |
| 4th Author's Name | Masahiro Iida |
| 4th Author's Affiliation | Kumamoto University(Kumamoto Univ.) |
| Date | 2022-09-08 |
| Paper # | RECONF2022-34 |
| Volume (vol) | vol.122 |
| Number (no) | RECONF-174 |
| Page | pp.pp.29-34(RECONF), |
| #Pages | 6 |
| Date of Issue | 2022-08-31 (RECONF) |