Compact Edge Vision Transformer with 86% Non-volatile Memory Bit Reduction by Percentile Clipping, Per Layer Quantization, and Quantization Aware Training

Ryuhei Yamaguchi

doi:https://doi.org/10.7567/SSDM.2023.K-8-03

Presentation information

Oral Presentation

02: Advanced and Emerging Memories / New Applications

[K-8] In-Memory and Unconventional Computing 2

Fri. Sep 8, 2023 3:30 PM - 4:30 PM Room K (234, Bldg. 2)

Session Chairs: Kazuyuki Kouno (Nuvoton Technology Corporation Japan), Xu Bai (NanoBridge Semiconductor, Inc.)

4:00 PM - 4:15 PM

[K-8-03] Compact Edge Vision Transformer with 86% Non-volatile Memory Bit Reduction by Percentile Clipping, Per Layer Quantization, and Quantization Aware Training

^○Ryuhei Yamaguchi¹, Ayumu Yamada¹, Naoko Misawa¹, Chihiro Matsui¹, Ken Takeuchi¹ (1. The University of Tokyo (Japan))

https://doi.org/10.7567/SSDM.2023.K-8-03

Abstract password authentication.
PA password is required to view abstracts. You can find the password in the "Advance Program".The "Advance Program" is handed out at the registration desk to registered participants.

Presentation information

[K-8] In-Memory and Unconventional Computing 2

[K-8-03] Compact Edge Vision Transformer with 86% Non-volatile Memory Bit Reduction by Percentile Clipping, Per Layer Quantization, and Quantization Aware Training

Password