Accuracy Tolerant Neural Networks Under Aggressive Power Optimization
Xiang-Xiu Wu1,a, Yi-Wen Hung1,b, Yung-Chih Chen2 and Shih-Chieh Chang1,3,c
1Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan
ajaubau999@gapp.nthu.edu.tw
bywhung@gapp.nthu.edu.tw
2Department of Computer Science & Engineering, Yuan Ze University, Taoyuan, Taiwan
ycchen.cse@saturn.yzu.edu.tw
3Electronic and Optoelectronic System Research Laboratories, ITRI, Hsinchu, Taiwan
cscchang@cs.nthu.edu.tw
ABSTRACT
With the success of deep learning, many neural network models have been proposed and applied to various applications. In several applications, the devices used to implement the complicated models have limited power resources, and thus aggressive optimization techniques are often applied for saving power. However, some optimization techniques, such as voltage scaling and multiple threshold voltages, may increase the probability of error occurrence due to slow signal propagation, which increases the path delay in a circuit and fails some input patterns. Although neural network models are considered to have some error tolerance, the prediction accuracy could be significantly affected, when there are a large number of errors. Thus, in this paper, we propose a scheme to mitigate the errors caused by slow signal propagation. Since the delay of multipliers dominates the critical path, we consider the patterns significantly altered by the slow signal propagation in a multiplier. We propose two methods, weight distribution and error-aware quantization to prevent the patterns from failure. Since we modify a neural network on the software side and it is unnecessary to re-design the hardware structure. The experimental results show that the proposed scheme is effective for several neural network models. It can improve the network accuracy by up to 27% under the consideration of slow signal propagation.
Keywords: Deep neural networks, timing violation, error mitigation, model optimization.
