Relationship between Complexity and Precision of Convolutional Neural Networks
Xiaolong Ke, Wenming Cao, Fangfang Lv · 2017
Convolutional neural networks (CNNs) have been successfully applied to the computer vision areas in recent years.However, these high performing CNNs generally involve intensive computation, which is unaffordable for many real-time applications.In this paper, we study the impact of four important network parameters Ƭ.Then we develop mathematical models to characterize the relationship and tradeoff between the complexity C and precision P of CNNs.Once the models C(Ƭ) and P(Ƭ) are obtained, we are able to perform complexity-precision optimization to minimize the CNN complexity while achieving the target precision level by selecting the optimal configuration of four network parameters Ƭ.