TY - GEN
T1 - Training and inference using approximate floating-point arithmetic for energy efficient spiking neural network processors
AU - Kwak, Myeongjin
AU - Lee, Jungwon
AU - Seo, Hyoju
AU - Sung, Mingyu
AU - Kim, Yongtae
N1 - Publisher Copyright:
© 2021 IEEE.
PY - 2021/1/31
Y1 - 2021/1/31
N2 - This paper presents a systematic analysis of spiking neural network (SNN) performance with reduced computation precisions using approximate adders. We propose an IEEE 754-based approximate floating-point adder that applies to the leaky integrate-and-fire (LIF) neuron-based SNN operation for both training and inference. The experimental results under a two-layer SNN for MNIST handwritten digit recognition application show that 4-bit exact mantissa adder with 19-bit approximation for lower-part OR adder (LOA), instead of 23-bit full-precision mantissa adder, can be exploited to maintain good classification accuracy. When adopted LOA as mantissa adder, it can achieve up to 74.1% and 96.5% of power and energy saving, respectively.
AB - This paper presents a systematic analysis of spiking neural network (SNN) performance with reduced computation precisions using approximate adders. We propose an IEEE 754-based approximate floating-point adder that applies to the leaky integrate-and-fire (LIF) neuron-based SNN operation for both training and inference. The experimental results under a two-layer SNN for MNIST handwritten digit recognition application show that 4-bit exact mantissa adder with 19-bit approximation for lower-part OR adder (LOA), instead of 23-bit full-precision mantissa adder, can be exploited to maintain good classification accuracy. When adopted LOA as mantissa adder, it can achieve up to 74.1% and 96.5% of power and energy saving, respectively.
KW - Approximate adder
KW - Floating-point arithmetic
KW - Leaky integrate-and-fire (LIF) neuron
KW - Spiking neural network (SNN)
UR - http://www.scopus.com/inward/record.url?scp=85102976345&partnerID=8YFLogxK
U2 - 10.1109/ICEIC51217.2021.9369724
DO - 10.1109/ICEIC51217.2021.9369724
M3 - Conference contribution
AN - SCOPUS:85102976345
T3 - 2021 International Conference on Electronics, Information, and Communication, ICEIC 2021
BT - 2021 International Conference on Electronics, Information, and Communication, ICEIC 2021
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2021 International Conference on Electronics, Information, and Communication, ICEIC 2021
Y2 - 31 January 2021 through 3 February 2021
ER -