TY - GEN
T1 - Design of an Energy-Efficient Accelerator for Training of Convolutional Neural Networks using Frequency-Domain Computation
AU - Ko, Jong Hwan
AU - Mudassar, Burhan
AU - Na, Taesik
AU - Mukhopadhyay, Saibal
N1 - Publisher Copyright:
© 2017 ACM.
PY - 2017/6/18
Y1 - 2017/6/18
N2 - Convolutional neural networks (CNNs) require high computation and memory demand for training. This paper presents the design of a frequency-domain accelerator for energy-efficient CNN training. With Fourier representations of parameters, we replace convolutions with simpler pointwise multiplications. To eliminate the Fourier transforms at every layer, we train the network entirely in the frequency domain using approximate frequency-domain nonlinear operations. We further reduce computation and memory requirements using sinc interpolation and Hermitian symmetry. The accelerator is designed and synthesized in 28nm CMOS, as well as prototyped in an FPGA. The simulation results show that the proposed accelerator significantly reduces training time and energy for a target recognition accuracy.
AB - Convolutional neural networks (CNNs) require high computation and memory demand for training. This paper presents the design of a frequency-domain accelerator for energy-efficient CNN training. With Fourier representations of parameters, we replace convolutions with simpler pointwise multiplications. To eliminate the Fourier transforms at every layer, we train the network entirely in the frequency domain using approximate frequency-domain nonlinear operations. We further reduce computation and memory requirements using sinc interpolation and Hermitian symmetry. The accelerator is designed and synthesized in 28nm CMOS, as well as prototyped in an FPGA. The simulation results show that the proposed accelerator significantly reduces training time and energy for a target recognition accuracy.
KW - convolutional neural network (CNN)
KW - frequency domain
KW - training
UR - https://www.scopus.com/pages/publications/85023641064
U2 - 10.1145/3061639.3062228
DO - 10.1145/3061639.3062228
M3 - Conference contribution
AN - SCOPUS:85023641064
T3 - Proceedings - Design Automation Conference
BT - Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 54th Annual Design Automation Conference, DAC 2017
Y2 - 18 June 2017 through 22 June 2017
ER -