Presentation | 2016-05-11 MERP-CNN: A Memory-Efficient Reconfigurable Processor for Convolutional Neural Networks Based on FPGA Xushen Han, Dajiang Zhou, Shinji Kimura, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Convolutional neural network has been paid so much attention in many intelligent applications especially image pattern recognition. Nowadays lots of high-throughput implementations are developed for acceleration of convolutional neural network, however, memory traffic has been the bottleneck if more resources are distributed on chip. In this paper we present MERP-CNN, a memory-efficient processor for both forward and backward propagation of convolutional neural networks. We implement our processor on Xilinx Vertex 7 FPGA platform. Our memory efficiency is 35.6% less than recent previous works. We also have a 279GFlops throughput for acceleration and not bad power performance. We take advantage of the multiboot specialty to keep top performance of our hardware resources. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | convolutional neural network / memory-efficient / multiboot |
Paper # | VLD2016-5 |
Date of Issue | 2016-05-04 (VLD) |
Conference Information | |
Committee | VLD / IPSJ-SLDM |
---|---|
Conference Date | 2016/5/11(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Kitakyushu International Conference Center |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | System Design, etc. |
Chair | Yusuke Matsunaga(Kyushu Univ.) / Masahiro Fukui(Ritsumeikan Univ.) |
Vice Chair | Takashi Takenana(NEC) |
Secretary | Takashi Takenana(Ritsumeikan Univ.) / (Fujitsu Labs.) |
Assistant | Ittetsu Taniguchi(Ritsumeikan Univ.) |
Paper Information | |
Registration To | Technical Committee on VLSI Design Technologies / Special Interest Group on System and LSI Design Methodology |
---|---|
Language | ENG |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | MERP-CNN: A Memory-Efficient Reconfigurable Processor for Convolutional Neural Networks Based on FPGA |
Sub Title (in English) | |
Keyword(1) | convolutional neural network |
Keyword(2) | memory-efficient |
Keyword(3) | multiboot |
1st Author's Name | Xushen Han |
1st Author's Affiliation | Waseda University(Waseda Univ.) |
2nd Author's Name | Dajiang Zhou |
2nd Author's Affiliation | Waseda University(Waseda Univ.) |
3rd Author's Name | Shinji Kimura |
3rd Author's Affiliation | Waseda University(Waseda Univ.) |
Date | 2016-05-11 |
Paper # | VLD2016-5 |
Volume (vol) | vol.116 |
Number (no) | VLD-21 |
Page | pp.pp.47-52(VLD), |
#Pages | 6 |
Date of Issue | 2016-05-04 (VLD) |