4. 様々な組み込み機器
4
By
Alexander
Glinz
-‐ photo
by
Alexander
Glinz
/
uploaded
by
Joadl,
CC
BY-‐SA
3.0at,https://commons.wikimedia.org/w/index.php?curid=29727889
GFDL,
https://ja.wikipedia.org/w/index.php?curid=248165
5. 組み込みにおけるNNの搭載
5
By
Alexander
Glinz
-‐ photo
by
Alexander
Glinz
/
uploaded
by
Joadl,
CC
BY-‐SA
3.0at,https://commons.wikimedia.org/w/index.php?curid=29727889
衝突回避システム
オンボード上でシステムを完結させたい
⾃自⽴立立制御ドローン
衛星リモート
センシング
26. BinaryConnect
M.Courbariaux, Y.Bengio, & David, J. P. (2015).
“Binaryconnect: Training deep neural networks with binary
weights during propagations”. In Advances in Neural
Information Processing Systems (pp. 3123-‐‑‒3131).
26
38. Binarized
Neural Network
M. Courbariaux et al. “Binarized Neural Networks:
Training Deep Neural Networks with Weights and
Activations Constrained to +1 or -‐‑‒1”. (2016): 12
June 2017.
38
55. XNOR-‐‑‒Net
M. Rastegari, V. Ordonez, J. Redmon, & A. Farhadi. (2016,
October). “Xnor-‐‑‒net: Imagenet classification using binary
convolutional neural networks”. In European Conference on
Computer Vision (pp. 525-‐‑‒542).
55
83. FINN
Y. Umuroglu, N. J. Fraser, G. Gambardella, M. Blott, P. Leong, M. Jahre, & K.
Vissers, (2017, February). “Finn: A framework for fast, scalable binarized
neural network inference”. In Proceedings of the 2017 ACM/SIGDA
International Symposium on Field-‐‑‒Programmable Gate Arrays (pp. 65-‐‑‒74).
ACM.
83
106. 乗算と加算の回路路⾯面積
106
Horowitz,
“Computing’s
Energy
Problem
(and
what
we
can
do
about
it)”,
ISSCC
2014
3495
137
0
500
1000
1500
2000
2500
3000
3500
32bit
乗算 32bit
加算
加算と乗算の回路路⾯面積[μm2]
乗算の約1/25
107. 乗算と加算の回路路⾯面積
107
Horowitz,
“Computing’s
Energy
Problem
(and
what
we
can
do
about
it)”,
ISSCC
2014
3495
137
0
500
1000
1500
2000
2500
3000
3500
32bit
乗算 32bit
加算
乗算と加算の回路路⾯面積[μm2]
同じ面積で24個
多く実装できる※
※回路路によって多少異異なる